CN114061766A - Multispectral reconstruction temperature measuring device and method in particle combustion process - Google Patents

Multispectral reconstruction temperature measuring device and method in particle combustion process Download PDF

Info

Publication number
CN114061766A
CN114061766A CN202111474356.5A CN202111474356A CN114061766A CN 114061766 A CN114061766 A CN 114061766A CN 202111474356 A CN202111474356 A CN 202111474356A CN 114061766 A CN114061766 A CN 114061766A
Authority
CN
China
Prior art keywords
radiation
image
particle
combustion process
multispectral
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111474356.5A
Other languages
Chinese (zh)
Inventor
杨斌
平力
樊荣
万代红
时志权
杨杨
王文松
闫亦菲
胡海航
倪虎
邱礼
车沈荣
刘哲昊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Minhang Collaborative Innovation Center Of Northwest University Of Technology
Shanghai Xinli Power Equipment Research Institute
University of Shanghai for Science and Technology
Original Assignee
Shanghai Minhang Collaborative Innovation Center Of Northwest University Of Technology
Shanghai Xinli Power Equipment Research Institute
University of Shanghai for Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shanghai Minhang Collaborative Innovation Center Of Northwest University Of Technology, Shanghai Xinli Power Equipment Research Institute, University of Shanghai for Science and Technology filed Critical Shanghai Minhang Collaborative Innovation Center Of Northwest University Of Technology
Priority to CN202111474356.5A priority Critical patent/CN114061766A/en
Publication of CN114061766A publication Critical patent/CN114061766A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01J—MEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J5/00—Radiation pyrometry, e.g. infrared or optical thermometry
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01J—MEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J5/00—Radiation pyrometry, e.g. infrared or optical thermometry
    • G01J2005/0077—Imaging

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Radiation Pyrometers (AREA)

Abstract

According to the multispectral reconstruction temperature measuring device and method for the particle combustion process, provided by the invention, direct imaging and spectral imaging technologies are combined, direct radiation imaging and radiation multispectral imaging in the particle combustion process are synchronously registered, the correlation between a radiation image and the radiation multispectral image in the particle combustion process is trained and excavated by adopting an artificial neural network, and the multispectral reconstruction is carried out on the radiation image in the particle combustion process based on a multispectral reconstruction and temperature inversion algorithm, so that the on-line measurement of the multi-component characteristic temperature distribution in the particle combustion process is realized, and the problems of low multi-phase component temperature measurement precision and low spatial resolution of the radiation spectral image in a radiation image method are solved. The multispectral reconstruction temperature measuring device for the particle combustion process is composed of a quartz glass plate, a beam splitter prism, a spectral imaging unit, a direct imaging unit, a synchronous trigger, an image processing unit and a control and signal cable.

Description

Multispectral reconstruction temperature measuring device and method in particle combustion process
Technical Field
The invention relates to the technical field of thermal measurement, in particular to a multispectral reconstruction temperature measuring device and method for a particle combustion process.
Background
In some industrial thermal energy combustion plants, solid particles of metals such as aluminum, magnesium, boron, etc. are typically added to the fuel to increase the heating value. These particle combustion parameters are closely related to the efficient utilization of fuel, wherein the temperature parameter is one of the important indexes directly reflecting the combustion state. The particle combustion temperature parameter accurate measurement is realized, and especially the particle combustion process temperature distribution measurement has important guiding effects on the aspects of researching the particle combustion mechanism, optimizing the design and combustion organization of combustion equipment, improving the energy utilization efficiency, reducing the pollutant emission, guaranteeing the safe operation of the equipment and the like.
The contact temperature measurement methods such as the thermocouple and the thermal resistor contact the sensor with the measured object, are single-point tests, have slow time response, and generally cannot cover the combustion temperature of the metal solid particles in the temperature measurement range. At present, with the development of Laser Spectroscopy, active Laser Spectroscopy technologies such as Planar Laser Induced Fluorescence (PLIF), Coherent Anti-stokes Raman Spectroscopy (CARS), Tunable Diode Laser Absorption Spectroscopy (TDLAS), and the like are rapidly developed, and the active Laser Spectroscopy technologies have the advantages of fast dynamic response, no interference to a combustion flow field, realization of online measurement, high space-time resolution, and the like, but for temperature measurement in a particle combustion process, the active Laser Spectroscopy technologies have the problems of strong spontaneous emission, scattering effect of particles on signals, and the like, and have low measurement accuracy.
Besides the advantage of non-contact measurement, the radiation image method does not need active light sources such as laser in the measurement system, and obtains the temperature distribution of the measured object by processing the radiation image signal of the measured object. However, the combustion temperature of the metal solid particles is high, solid, liquid and gas multi-phase components exist in the combustion process, and the radiation characteristics of the multi-phase components are different. The radiation image method cannot determine the radiation characteristic rule of each component in the particle combustion process, so that the obtained multi-phase component temperature distribution difference in the particle combustion process is large.
With the progress of the spectral imaging technology and the development of the artificial intelligence algorithm, more spectral information and spatial information of the combustion particles are detected, and parameter analysis and model solution of various complex algorithms become possible, so that on the basis of ensuring the temperature measurement spatial resolution of the combustion particles, more spectral information on the spatial position is obtained, the multispectral reconstruction temperature measurement in the particle combustion process is realized, and the temperature measurement spatial resolution in the particle combustion process can be effectively improved.
Disclosure of Invention
The invention mainly solves the technical problem of providing a particle combustion process multispectral reconstruction temperature measuring device and a particle combustion process multispectral reconstruction temperature measuring method, which combine direct imaging and spectral imaging technologies, synchronize direct radiation imaging and radiation multispectral imaging in a particle combustion process, adopt artificial neural network training to excavate the correlation between a radiation image and the radiation multispectral image in the particle combustion process, perform multispectral reconstruction on the radiation image in the particle combustion process based on a multispectral reconstruction and temperature inversion algorithm, realize the online measurement of multicomponent characteristic temperature distribution in the particle combustion process, and solve the problems of low temperature measuring precision of the multicomponent component and low spatial resolution of the radiation spectral image in a radiation image method.
In order to solve the technical problem, the invention provides a multispectral reconstruction temperature measuring device for a particle combustion process, which is characterized by comprising a quartz glass plate, a beam splitter prism, a spectral imaging unit, a direct imaging unit, a synchronous trigger, an image processing unit and a control and signal cable, wherein particle radiation light is split by the beam splitter prism through the quartz glass plate, one beam of light enters the spectral imaging unit to obtain a particle combustion radiation spectral image with high spectral resolution, and the other beam of light enters the direct imaging unit to obtain a particle combustion radiation spectral image with high spatial resolution; the quartz glass plate is used for preventing combustion particles from polluting optical components; the synchronous trigger is used for synchronizing the simultaneous work of the spectral imaging unit and the direct imaging unit; the image processing unit is used for synchronizing direct radiation imaging and radiation multispectral imaging registration in the particle combustion process, adopting an artificial neural network to train and excavate the correlation between the radiation image and the radiation multispectral image in the particle combustion process, and carrying out multispectral reconstruction on the radiation image in the particle combustion process based on a multispectral reconstruction and temperature inversion algorithm so as to realize online measurement of multicomponent characteristic temperature distribution in the particle combustion process.
The spectral imaging unit is characterized by comprising a spectral imaging attenuation sheet, a spectral imaging lens and a spectral camera, wherein a beam of light of the particle radiation light is split by the light splitting prism and enters the spectral imaging attenuation sheet to be attenuated, and then the beam of light is adjusted by the spectral imaging lens and enters the spectral camera to be imaged to obtain a radiation multispectral image of the particle combustion process.
The direct imaging unit is characterized by comprising a direct imaging attenuation sheet, a direct imaging lens and an industrial camera, wherein after being split by the beam splitter prism, the particle radiation light enters the direct imaging attenuation sheet for attenuation, is adjusted by the direct imaging lens and enters the industrial camera for imaging, and a particle combustion process radiation image is obtained.
The synchronous trigger is characterized in that the synchronous trigger is connected with a spectrum camera of the spectrum imaging unit and an industrial camera of the direct imaging unit through a control cable and is used for synchronously triggering the spectrum imaging unit and the direct imaging unit to work simultaneously.
The image processing part is characterized in that the image processing part is connected with a spectrum camera of the spectrum imaging unit and an industrial camera of the direct imaging unit through signal cables, the correlation between the radiation image and the radiation multispectral image in the particle combustion process is trained and excavated by adopting an artificial neural network through obtaining the synchronous image of direct radiation imaging and radiation multispectral imaging registration in the particle combustion process, and the multispectral reconstruction is carried out on the radiation image in the particle combustion process based on a multispectral reconstruction and temperature inversion algorithm, so that the online measurement of the multicomponent characteristic temperature distribution in the particle combustion process is realized.
The spectral camera is characterized by comprising a condensing lens, an optical slit, an adjusting lens, a grating, an imaging lens, a micro-array lens, a photoelectric detection array, a digital image acquisition unit and an image processor. The particle radiation optical signal focusing imaging device comprises a particle radiation optical signal focusing imaging position plane, a focusing lens and a focusing lens, wherein the focusing lens is used for adjusting the particle radiation optical signal focusing imaging position plane; the optical slits are used for reducing the influence of ambient light; the adjusting lens is used for adjusting the particle image radiation light to enter the grating in parallel; the grating is used for decomposing optical signals with different wavelengths; the imaging lens is used for adjusting the wavelength splitting signal to enter the micro-array lens; the micro-array lens is used for converging light to enter the photoelectric detection array; the photoelectric detection array is used for converting optical signals into electric signals; the digital image acquisition unit is used for converting the received photoelectric detection array electric signal into a digital image signal; the image processor is used for processing the received digital image signals to obtain particle combustion radiation spectrum images.
Furthermore, the synchronous trigger is a signal generator, is a double-channel synchronous trigger and can select a pulse signal as a trigger signal.
Further, the image processing unit may be a computer.
Further, the beam splitter prism is a 50% -50% beam splitter prism.
Further, the attenuation parameters of the spectral imaging attenuation sheet 11 and the direct imaging attenuation sheet 21 are determined according to the particle burning radiation optical signal and the corresponding camera response range.
The invention also provides a particle combustion process multispectral reconstruction temperature measurement method, which is characterized by mainly comprising two processes of direct radiation imaging and radiation multispectral imaging registration synchronization in the particle combustion process, adopting artificial neural network training to excavate a particle combustion process radiation image and a radiation multispectral image, and carrying out multispectral reconstruction temperature measurement on the particle combustion process radiation image based on a multispectral reconstruction and temperature inversion algorithm, and specifically comprising the following steps of:
s1: a particle combustion process multispectral reconstruction temperature measuring device is adopted, and a radiation image and a radiation multispectral image in the particle combustion process are synchronously obtained through direct radiation imaging and radiation multispectral imaging registration in the particle combustion process;
s2: normalizing the grain size of the particle combustion image;
s3: angle correction and translation correction of synchronous particle combustion images;
s4: target identification of particle combustion images;
s5: establishing an artificial neural network model of the spectrum dimension and the image dimension;
s6: training an artificial neural network model based on an error back propagation algorithm;
s7: optimizing the number of hidden layers and the number of neurons of the artificial neural network;
s8: and performing multispectral reconstruction on the combustion particle radiation image and inverting the temperature of the combustion particle radiation image.
The particle burning process direct radiation imaging and radiation multi-spectral imaging registration synchronization process comprises the steps of S2, S3 and S4.
The process of training the radiation image and the radiation multispectral image of the burning process of the excavation particles by adopting the artificial neural network comprises the steps of S5, S6 and S7.
The grain combustion image texture size normalization is to perform texture size normalization on response images of an industrial camera and a spectral camera sensor. Generally, the pixel resolution of an industrial camera image is higher than that of lightThe process of spectral camera image, and therefore texture size normalization, is also a process of degrading the sharpness of the radiation image to the sharpness of the radiation multispectral image. The acquired radiation image can be represented in a matrix formX M N××3WhereinM×NFor its spatial resolution, the radiation multispectral image matrix is represented asY m n l××With a spatial resolution ofm×n，lIs the number of bands. Need to be provided withXResolution ofM×NDegenerating tom×nThe pixels of the two camera image data sets can be subjected to model training, and the factors of the resolution ratio and the pixel area of the sensor and the degradation coefficient of the radiation image are considered in the processCExpressed as:
Figure 927206DEST_PATH_IMAGE001
wherein,S X is the pixel area of an industrial camera sensor,S Y is the pixel area of the spectral camera sensor, and the matrixX M N××3Degenerating toX m n××3The texture size normalization of the radiation image and the radiation multispectral image can be realized, so that each corresponding pixel point in the radiation image displays the same texture position of the combustion particles.
The angle correction and the translation correction of the synchronous particle combustion image are to solve the problem that the images of the same object detected by two camera sensors may have angle deviation and translation amount after a particle combustion radiation signal passes through a beam splitting prism, a backlight method is adopted to shoot a calibration plate image, and a cross-correlation template matching analysis method is utilized to calculate and obtain the translation amount and the rotation angle between the two images. The cross-correlation analysis of template matching is to define the radiation image and radiation multispectral image after graying as image template matrixXAndYimages are matched by analyzing cross-correlation coefficients between templates. Template matrixXThrough displacement and rotation change to obtainX i j θ,,The variation formula is as follows:
Figure 723124DEST_PATH_IMAGE002
wherein,m、nrepresenting a template matrixXThe column-row value of (a),iandjas a templateXThe amount of translation in the parallel and vertical directions,θis an anticlockwise rotating angle and obtains the template after conversionX i j θ,,。
ComputingX i j θ,,AndYcross correlation coefficient betweenR(i, j, θ) When the cross-correlation coefficient reaches a maximum value, corresponding toi, j, θ) Namely the templateXAndYthe amount of translation and the angle of rotation therebetween.
Figure 476316DEST_PATH_IMAGE003
Therefore, the optimal solution of the translation amount and the angle is obtained through the cross-correlation analysis of the calibration plate images.
The particle combustion image target identification adopts a Canny edge processing method, the gradient amplitude and the direction of the gray level image are calculated, the point with the maximum local gradient is reserved by adopting a non-maximum inhibition method for the gradient value, and finally the image is thresholded.
The establishing of the artificial neural network model of the spectrum dimension and the image dimension is based on the artificial neural network to establish a mathematical relationship model between the direct radiation imaging and the radiation multispectral imaging in the particle combustion process on the basis of synchronous registration. In the process of multispectral reconstruction of the radiation image, response signals of the radiation image at each pixel point (x r , x g , x b ) And (3) as an input layer variable of the neural network, sending the variable into the model, wherein the relational expression of each neuron model in the first layer of the hidden layer is as follows:
Figure 357684DEST_PATH_IMAGE004
in the formula,x j response intensity on R, G, B wave band of each pixel point in the radiation image;w i j,to representx j The weight coefficient of (a);b i is composed ofx j A bias threshold of (a);kis composed ofb i Andw i j,the dimension of (a) is the number of neuron units in the hidden layer; the weight coefficients and bias thresholds of the neuron elements represent a linear processing of the data by the network,u i the sum of the two is obtained.
The artificial neural network model trained based on the error back propagation algorithm adopts a nonlinear function in the neural network model to utilize an excitation functionϕ(u) Expressing:
Figure 556322DEST_PATH_IMAGE005
will be provided withϕ(u) As the output value of the first layer of the hidden layer, the output value is sent to the next layer of the neural network, if the weight coefficient of the current layer is weightedw i j,And paranoia thresholdb i Is defined as a weight matrixWAnd a bias matrixBThe input value and the output value of the layer areXAndYa matrix, which can be expressed as:
Figure 675588DEST_PATH_IMAGE006
the specific matrix operation is as follows:
Figure 599682DEST_PATH_IMAGE007
the resulting matrixYSending to the next layer of the network model until reaching the output layer, and outputting the result (z 1, z 2, …, z m ) And realizing the feed-forward propagation of data in the artificial neural network, wherein,mis the dimension of the output layer.Calculating the error between the expected output value and the network output value by the artificial neural network model by using a loss function, and taking the radiation multispectral image corresponding to the radiation image as the expected output value of the neural network model (z’ 1, z’ 2, …, z’ m ). The training results of the network were evaluated using Mean Squared Error (MSE) as a loss function:
Figure 968346DEST_PATH_IMAGE008
MSE is the error of the prediction result, and the weight coefficient of each layer in the network model is adjusted by using a gradient descent method according to the value of the errorWAnd bias thresholdBDetermining the adjustment degree of the parameters in the network model by solving the partial derivative of the mean square error to each parameter in the output layer:
Figure 472140DEST_PATH_IMAGE009
Figure 475606DEST_PATH_IMAGE010
wherein,ηthe step length is adjusted, namely the learning rate of the neural network model is obtained.
The artificial neural network implicit layer number and neuron number optimization utilizes a gradient descent method to continuously adjust the weight and bias of the neuron units from an output layer to an input layer, so that the MSE of the output value and the expected output value of the neural network is continuously reduced, reverse transmission of errors and iteration of an algorithm are realized until the MSE between the output value and the expected value of the network model reaches a preset target value, and training of the neural network is completed.
The multispectral reconstruction of the combustion particle radiation image is realized by adopting the radiation image after pixel degradation and texture information of combustion particles in the radiation multispectral radiation image to correspond one to one according to pixel points, and establishing a mathematical model through the mapping relation between the response intensity of the radiation image and the multispectral response intensity.
The multi-spectral temperature inversion method for the combustion particle radiation image is characterized by comprising the following steps of:
s81: acquiring a particle combustion radiation spectrum image by adopting a particle combustion process multispectral reconstruction temperature measuring device, and performing multispectral reconstruction;
s82: identifying the multiphase components in the particle combustion process and decomposing an image area by utilizing a multiphase component identification and image decomposition algorithm in the particle combustion process;
s83: processing the same-phase component image area in the particle combustion process, and obtaining the radiation characteristic of the phase component through the radiation spectrum of each pixel point by using a radiation characteristic algorithm based on the radiation spectrum;
s84: processing the radiation spectrum of each pixel point in the same-phase component image area in the particle combustion process, and obtaining a temperature value of the pixel point by using a radiation spectrum temperature inversion algorithm;
s85: based on the temperature value result of each pixel point, the image of the multi-phase component temperature result in the particle combustion process is sorted and displayed by utilizing a multi-phase component temperature result image integration algorithm in the particle combustion process.
The particle combustion process multi-phase component identification and image decomposition algorithm is characterized in that different phase component radiation images are identified according to a multi-band spectral response value clustering analysis algorithm of each pixel point of a combustion particle radiation image, and the images are decomposed into imaging backgrounds and particle combustion phase image areas.
The clustering analysis algorithm for the multiband spectral response values of all pixel points of the combustion particle radiation image is characterized in that the clustering analysis algorithm randomly selects the multiband spectral response values of all pixel points of the combustion particle radiation image from a data set containing the multiband spectral response valueskTaking the data samples as initial clustering centers, and counting each spectrum sample andkthe distance of each initial clustering center, all the spectral data are divided into the categories represented by the clustering centers with the closest distances to the initial clustering centers, and the spectral data are updated according to the mean value of the spectral samples in the newly generated categorieskAnd (4) clustering centers. If the variation of the clustering central value exceeds the set threshold value within the adjacent iteration timesThen, classifying all the data samples again according to the new clustering center; and if the change of the clustering central value in the adjacent iteration times is smaller than a specified threshold value, the algorithm converges and a clustering result is output, so that the multi-phase component identification and the image decomposition in the particle combustion process are realized.
The clustering analysis algorithm for multiband spectral response values of all pixel points of the combustion particle radiation image is characterized in that the processing flow of the algorithm is as follows:
(1) selecting a clustered original data set;
(2) from which random selection is madekUsing the data sample as initial clustering centerz 1, z 2, …, z k ；
(3) Calculating all data sample data one by one tokThe distance between each condensation point (usually Euclidean distance is used as the distance from the sample to the cluster center), and the distance is determined according to the size of the distancenEach sample (or variable) being divided intokClass, Euclidean distance calculation formula is as follows:
Figure 570601DEST_PATH_IMAGE011
wherein,x i is a samplexTo (1) aiThe value of a variable of the individual variables,y i is a sampleyIs/are as followsiThe variable values of the individual variables. If the distance from the data sample to the original class is the shortest, the data sample is still in the original class, otherwise, the data sample is moved to the class which is the shortest;
(4) computingkAnd (3) if the clustering center of each type of data in the class is not coincident with the initial clustering center, taking the clustering center as a new clustering center, repeating the step (3) until all the spectrum samples cannot move or each clustering center is not changed, and terminating the calculation process.
The radiation characteristic algorithm based on the radiation spectrum is characterized in that the radiation spectrum of each pixel point is processed by the algorithm, the Planck radiation law is adopted to describe the radiation spectrum of an object, the spectral radiation force is distributed along with the wavelength,spectral radiation force of black body on each wave bandE b Can be expressed as:
Figure 895403DEST_PATH_IMAGE012
wherein,E b is the radiation intensity of black body in each wave band, W.m-2；C 1AndC 2respectively a planck first radiation constant and a second radiation constant,C 1=3.7419×10-16W·m-2、C 2=1.4388×10-2m·K；λis the wavelength, m;Tis the temperature, K;
the radiance of the material is constructed by adopting the following polynomial functionε(λ)：
Figure 937308DEST_PATH_IMAGE013
Wherein,a i is a constant of a polynomial and is,nis the fitting order;
and determining a wave band range meeting the radiation gray property according to a radiation gray property judgment method by obtaining the phase state radiance model to which each pixel point belongs.
The radiation grayness judgment method is characterized in that the ratio of monochromatic response intensities on two wavelengths is calculated:
Figure 562324DEST_PATH_IMAGE014
wherein,ε(λ)、ε(λ+Δλ) Respectively the burning particles at wavelengthλ、λ+ΔλThe radiance of (c). When the wavelength changes by an amount of ΔλSufficiently small, the temperature value can be calculated from the ratio of the radiation intensities of the two wavelengths:
Figure 828221DEST_PATH_IMAGE015
after the temperature value is obtained, the radiance distribution in the wavelength range can be calculated according to the response intensity of the measured object and the response intensity of the black body at the same temperature:
Figure 416152DEST_PATH_IMAGE016
wherein,I b (λ) Is the response value of the blackbody radiation spectrum;
according to the change condition of the radiance distribution along with the wavelength, whether the combustion particles meet the radiance of the ash body can be judged. If the emissivity changes more smoothly with wavelength, the soot body radiation characteristics of the combustion particles in this band are better, and if the change is more dramatic, they should not be considered soot bodies in this band. Analyzing the fluctuation of radiance distribution by using relative variance, and obtaining average radiance in a certain waveband rangeε(a)：
Figure 58486DEST_PATH_IMAGE017
Wherein,mis the wave number;
the variance of radiance over this band can be expressed as:
Figure 538009DEST_PATH_IMAGE018
with respect to the emissivity variance, it is generally considered that when it is less than 5%, the radiation characteristic of the combustion particles in this wavelength band satisfies the radiation characteristic of the soot body.
The radiation spectrum temperature inversion algorithm is characterized in that the algorithm obtains the temperature value of each pixel point based on a Planck's law parameter fitting algorithm according to the selected spectrum data in the wave band range meeting radiation grayness.
The Planck's law parameter fitting algorithm is characterized in that through obtaining photoelectric response values of all wave bands of all pixel points of a radiation image, response intensity can be searched by utilizing a least square method parameter fitting algorithmI(λ,T) AndE(λ,T) Optimal matching parameter betweenεAndT，the temperature is obtained.
The least square method parameter fitting algorithm is characterized in that a function model is established by the least square parameter fitting principleF(ε, T) Analyzing the optimal solution of the temperature and emissivity parameters of the burning particles:
Figure 912490DEST_PATH_IMAGE019
in the formula,Iin order to obtain a spectral response value by measurement,kthe obtained response coefficient is calibrated through photoelectric response.
The particle combustion process multi-phase component temperature result image integration algorithm is characterized in that the algorithm integrates the multi-phase component temperature result of the particle combustion process according to the coordinates of all pixel points and the temperature measurement result to form a particle combustion process multi-phase component temperature distribution image.
The invention relates to a multispectral reconstruction temperature measuring device and a multispectral reconstruction temperature measuring method in a particle combustion process, which have the following functions and effects:
(1) combining direct imaging and spectral imaging technologies, synchronizing direct radiation imaging and radiation multispectral imaging in a particle combustion process through registration, training and excavating the correlation between a radiation image and a radiation multispectral image in the particle combustion process by adopting an artificial neural network, and performing multispectral reconstruction on the radiation image in the particle combustion process based on multispectral reconstruction and a temperature inversion algorithm to realize online measurement of multi-component characteristic temperature distribution in the particle combustion process and solve the problems of low multi-phase component temperature measurement precision and low spatial resolution of the radiation spectrum image in a radiation image method;
(2) the method has the advantages that the multi-phase component spectrum in the particle combustion process is imaged in a micro-grating array spectroscopic imaging mode, the spectrum and the image information in the particle combustion process can be obtained on line at the same time, the spectrum and the image information in the particle combustion process are subjected to temperature distribution processing, and the problem that the multi-phase component temperature distribution in the particle combustion process cannot be accurately measured due to different radiation characteristics of different phase components can be solved;
(3) by utilizing a multi-phase component identification and image decomposition algorithm in the particle combustion process and according to the clustering analysis of multi-band spectral response values of all pixel points of a combustion particle radiation image, the multi-phase component in the particle combustion process can be identified, and the spectral images are subjected to region decomposition and are processed one by one according to different phase states;
(4) based on the radiation spectrum of each pixel point in the same-phase component image area in the particle combustion process, a phase state radiance model to which each pixel point belongs can be obtained through polynomial function fitting, and the wave band range meeting radiation gray can be determined according to a radiation gray judgment method, so that the temperature inversion is carried out on the multiple-phase components in the particle combustion process one by one;
(5) and obtaining the temperature value of each pixel point based on the Planck's law parameter fitting algorithm according to the selected spectral data in the wave band range meeting the radiation gray property, and integrating the temperature results of the multiphase components in the particle combustion process according to the coordinates of each pixel point and the temperature measurement results to form a multiphase component temperature distribution image in the particle combustion process.
Drawings
FIG. 1 is a schematic structural diagram of a multispectral reconstruction temperature measuring device for a particle combustion process.
Fig. 2 is a schematic diagram of the structure of the spectral camera.
FIG. 3 is a process flow diagram of a method for multi-spectral reconstruction thermometry of a particulate combustion process.
FIG. 4 is a schematic diagram of an artificial neural network training model for a particle combustion process radiation image and a radiation multispectral image.
FIG. 5 is a flow chart of a multi-spectral temperature inversion algorithm for a combustion particle radiation image.
Detailed Description
The following detailed description of the preferred embodiments of the present invention, taken in conjunction with the accompanying drawings, will make the advantages and features of the invention easier to understand by those skilled in the art, and thus will clearly and clearly define the scope of the invention.
Referring to fig. 1-5, an embodiment of the present invention includes:
as shown in fig. 1, an embodiment of the present invention provides a particle combustion process multispectral reconstruction temperature measurement apparatus, which includes a quartz glass plate 01, a beam splitter prism 02, a spectral imaging unit 10, a direct imaging unit 20, a synchronous trigger 3, an image processing unit 4, and control and signal cables 14, 15, 24, and 25, wherein a particle 0 radiation beam 100 is split by the beam splitter prism 02 via the quartz glass plate 01, one beam of light 101 enters the spectral imaging unit 10 to obtain a particle combustion radiation spectral image with high spectral resolution, and the other beam of light 102 enters the direct imaging unit 20 to obtain a particle combustion radiation spectral image with high spatial resolution; the quartz glass plate 01 is used for preventing combustion particles from polluting optical components; the synchronous trigger 3 is used for synchronizing the simultaneous work of the spectral imaging unit 10 and the direct imaging unit 20; the image processing unit 4 is used for synchronizing direct radiation imaging and radiation multispectral imaging registration in the particle combustion process, training and excavating the correlation between the radiation image and the radiation multispectral image in the particle combustion process by adopting an artificial neural network, and performing multispectral reconstruction on the radiation image in the particle combustion process based on a multispectral reconstruction and temperature inversion algorithm to realize online measurement of multicomponent characteristic temperature distribution in the particle combustion process.
The spectral imaging unit 10 comprises a spectral imaging attenuation sheet 11, a spectral imaging lens 12 and a spectral camera 13, wherein a beam of light 101 of the particle 0 radiation light 100 is split by the beam splitter prism 02 and enters the spectral imaging attenuation sheet 11 to be attenuated, and then the beam is adjusted by the spectral imaging lens 12 and enters the spectral camera 13 to be imaged, so that a radiation multispectral image of the particle combustion process is obtained.
The direct imaging unit 20 comprises a direct imaging attenuation sheet 21, a direct imaging lens 22 and an industrial camera 23, wherein after being split by the beam splitter prism 02, the particle 0 radiation light 100 enters the direct imaging attenuation sheet 21 for attenuation, is adjusted by the direct imaging lens 22 and enters the industrial camera 23 for imaging, and a radiation image in the particle combustion process is obtained.
The synchronous trigger 3 is connected with the spectrum camera 13 of the spectrum imaging unit 10 and the industrial camera 23 of the direct imaging unit 20 through control cables 14 and 24, and is used for synchronously triggering the spectrum imaging unit 10 and the direct imaging unit 20 to work simultaneously.
The image processing unit 4 is connected with the spectrum camera 13 of the spectrum imaging unit 10 and the industrial camera 23 of the direct imaging unit 20 through signal cables 15 and 25, and by obtaining a synchronous image of direct radiation imaging and radiation multispectral imaging registration in the particle combustion process, the correlation between the radiation image and the radiation multispectral image in the particle combustion process is trained and excavated by adopting an artificial neural network, and the multispectral reconstruction is carried out on the radiation image in the particle combustion process based on a multispectral reconstruction and temperature inversion algorithm, so that the online measurement of the multicomponent characteristic temperature distribution in the particle combustion process is realized.
As shown in fig. 2, the spectrum camera 13 is composed of a condenser lens 131, an optical slit 132, an adjusting lens 133, a grating 134, an imaging lens 135, a micro-array lens 136, a photo-detection array 137, a digital image acquisition unit 138, and an image processor 139. The condenser lens 131 is used for adjusting the condensing imaging position plane of the optical signal radiated by the particle 001 (gas phase component 002, solid phase component 003); the optical slits 132 are used to reduce the ambient light effect; the adjusting lens 133 is used for adjusting the parallel entering of the particle image radiation light into the grating 134; the grating 134 is used for decomposing optical signals with different wavelengths; the imaging lens 135 is used to condition the wavelength-division optical signal to enter the micro-array lens 136; the micro-array lens 136 is used for converging light into the photoelectric detection array 137; the photo detection array 137 is used for converting optical signals into electrical signals; the digital image acquisition unit 138 is configured to convert the received electrical signal of the photoelectric detection array into a digital image signal; the image processor 139 is configured to process the received digital image signal to obtain a particle combustion radiation spectrum image.
Further, the synchronous trigger 3 is a signal generator which is a dual-channel synchronous trigger, and can select a pulse signal as a trigger signal.
Further, the image processing unit 4 may be a computer.
Further, the beam splitter prism 02 is a 50% -50% beam splitter prism.
Further, the attenuation parameters of the spectral imaging attenuation sheet 11 and the direct imaging attenuation sheet 21 are determined according to the particle burning radiation optical signal and the corresponding camera response range.
Another aspect of an embodiment of the present invention provides a method for multi-spectral reconstruction thermometry in a particle combustion process,
the particle combustion process radiation image multi-spectral reconstruction temperature measurement method mainly comprises the steps of synchronizing direct radiation imaging and radiation multi-spectral imaging in a particle combustion process, training and excavating a particle combustion process radiation image and a radiation multi-spectral image by adopting an artificial neural network, and performing multi-spectral reconstruction temperature measurement on the particle combustion process radiation image based on a multi-spectral reconstruction and temperature inversion algorithm, and specifically comprises the following steps as shown in figure 3:
s1: a particle combustion process multispectral reconstruction temperature measuring device is adopted, and a radiation image and a radiation multispectral image in the particle combustion process are synchronously obtained through direct radiation imaging and radiation multispectral imaging registration in the particle combustion process;
s2: normalizing the grain size of the particle combustion image;
s3: angle correction and translation correction of synchronous particle combustion images;
s4: target identification of particle combustion images;
s5: establishing an artificial neural network model of the spectrum dimension and the image dimension;
s6: training an artificial neural network model based on an error back propagation algorithm;
s7: optimizing the number of hidden layers and the number of neurons of the artificial neural network;
s8: and performing multispectral reconstruction on the combustion particle radiation image and inverting the temperature of the combustion particle radiation image.
The particle burning process direct radiation imaging and radiation multi-spectral imaging registration synchronization process comprises the steps of S2, S3 and S4.
The process of training the radiation image and the radiation multispectral image of the burning process of the excavation particles by adopting the artificial neural network comprises the steps of S5, S6 and S7.
The grain combustion image texture size normalization is to perform texture size normalization on response images of an industrial camera and a spectral camera sensor. In general, the pixel resolution of an industrial camera image is higher than that of a spectral camera image, and therefore, the process of texture size normalization is also a process of degrading the sharpness of a radiation image to the sharpness of a radiation multispectral image. The acquired radiation image can be represented in a matrix formX M N××3WhereinM×NFor its spatial resolution, the radiation multispectral image matrix is represented asY m n l××With a spatial resolution ofm×n，lIs the number of bands. Need to be provided withXResolution ofM×NDegenerating tom×nThe pixels of the two camera image data sets can be subjected to model training, and the factors of the resolution ratio and the pixel area of the sensor and the degradation coefficient of the radiation image are considered in the processCExpressed as:
Figure 477463DEST_PATH_IMAGE001
wherein,S X is the pixel area of an industrial camera sensor,S Y is the pixel area of the spectral camera sensor, and the matrixX M N××3Degenerating toX m n××3The texture size normalization of the radiation image and the radiation multispectral image can be realized, so that each corresponding pixel point in the radiation image displays the same texture position of the combustion particles.
The angle correction and the translation correction of the synchronous particle combustion image are to solve the problem that the images of the same object detected by two camera sensors may have angle deviation and translation amount after a particle combustion radiation signal passes through a beam splitting prism, a backlight method is adopted to shoot a calibration plate image, and a cross-correlation template matching analysis method is utilized to calculate and obtain the translation amount and the rotation angle between the two images. The cross-correlation analysis of template matching is to define the radiation image and radiation multispectral image after graying as image template matrixXAndYimages are matched by analyzing cross-correlation coefficients between templates. Template matrixXThrough displacement and rotation change to obtainX i j θ,,The variation formula is as follows:
Figure 392329DEST_PATH_IMAGE002
wherein,m、nrepresenting a template matrixXThe column-row value of (a),iandjas a templateXThe amount of translation in the parallel and vertical directions,θis an anticlockwise rotating angle and obtains the template after conversionX i j θ,,。
ComputingX i j θ,,AndYcross correlation coefficient betweenR(i, j, θ) When the cross-correlation coefficient reaches a maximum value, corresponding toi, j, θ) Namely the templateXAndYthe amount of translation and the angle of rotation therebetween.
Figure 490473DEST_PATH_IMAGE003
Therefore, the optimal solution of the translation amount and the angle is obtained through the cross-correlation analysis of the calibration plate images.
The particle combustion image target identification adopts a Canny edge processing method, the gradient amplitude and the direction of the gray level image are calculated, the point with the maximum local gradient is reserved by adopting a non-maximum inhibition method for the gradient value, and finally the image is thresholded.
The establishing of the artificial neural network model of the spectrum dimension and the image dimension is based on the artificial neural network to establish a mathematical relationship model between the direct radiation imaging and the radiation multispectral imaging in the particle combustion process on the basis of synchronous registration. In the process of multispectral reconstruction of the radiation image, response signals of the radiation image at each pixel point (x r , x g , x b ) And (3) as an input layer variable of the neural network, sending the variable into the model, wherein the relational expression of each neuron model in the first layer of the hidden layer is as follows:
Figure 98172DEST_PATH_IMAGE004
in the formula,x j response intensity on R, G, B wave band of each pixel point in the radiation image;w i j,to representx j The weight coefficient of (a);b i is composed ofx j A bias threshold of (a);kis composed ofb i Andw i j,the dimension of (a) is the number of neuron units in the hidden layer; the weight coefficients and bias thresholds of the neuron elements represent a linear processing of the data by the network,u i the sum of the two is obtained.
The artificial neural network model trained based on the error back propagation algorithm is shown in FIG. 4, and a nonlinear function in the neural network model is adopted to utilize an excitation functionϕ(u) Expressing:
Figure 619283DEST_PATH_IMAGE005
will be provided withϕ(u) As the output value of the first layer of the hidden layer, the output value is sent to the next layer of the neural network, if the weight coefficient of the current layer is weightedw i j,And paranoia thresholdb i Is defined as a weight matrixWAnd a bias matrixBThe input value and the output value of the layer areXAndYa matrix, which can be expressed as:
Figure 337841DEST_PATH_IMAGE006
the specific matrix operation is as follows:
Figure 791956DEST_PATH_IMAGE007
the resulting matrixYSending to the next layer of the network model until reaching the output layer, and outputting the result (z 1, z 2, …, z m ) And realizing the feed-forward propagation of data in the artificial neural network, wherein,mis the dimension of the output layer. Calculating the error between the expected output value and the network output value by the artificial neural network model by using a loss function, and taking the radiation multispectral image corresponding to the radiation image as the expected output value of the neural network model (z’ 1, z’ 2, …, z’ m ). The training results of the network were evaluated using Mean Squared Error (MSE) as a loss function:
Figure 570556DEST_PATH_IMAGE008
MSE is the error of the prediction result, and the weight coefficient of each layer in the network model is adjusted by using a gradient descent method according to the value of the errorWAnd bias thresholdBDetermining the adjustment degree of the parameters in the network model by solving the partial derivative of the mean square error to each parameter in the output layer:
Figure 110122DEST_PATH_IMAGE009
Figure 130905DEST_PATH_IMAGE010
wherein,ηthe step length is adjusted, namely the learning rate of the neural network model is obtained.
The artificial neural network implicit layer number and neuron number optimization utilizes a gradient descent method to continuously adjust the weight and bias of the neuron units from an output layer to an input layer, so that the MSE of the output value and the expected output value of the neural network is continuously reduced, reverse transmission of errors and iteration of an algorithm are realized until the MSE between the output value and the expected value of the network model reaches a preset target value, and training of the neural network is completed.
The multispectral reconstruction of the combustion particle radiation image is realized by adopting the radiation image after pixel degradation and texture information of combustion particles in the radiation multispectral radiation image to correspond one to one according to pixel points, and establishing a mathematical model through the mapping relation between the response intensity of the radiation image and the multispectral response intensity.
The multi-spectral temperature inversion of the combustion particle radiation image, as shown in fig. 5, comprises the following steps:
s81: acquiring a particle combustion radiation spectrum image by adopting a particle combustion process multispectral reconstruction temperature measuring device, and performing multispectral reconstruction;
s82: identifying the multiphase components in the particle combustion process and decomposing an image area by utilizing a multiphase component identification and image decomposition algorithm in the particle combustion process;
s83: processing the same-phase component image area in the particle combustion process, and obtaining the radiation characteristic of the phase component through the radiation spectrum of each pixel point by using a radiation characteristic algorithm based on the radiation spectrum;
s84: processing the radiation spectrum of each pixel point in the same-phase component image area in the particle combustion process, and obtaining a temperature value of the pixel point by using a radiation spectrum temperature inversion algorithm;
s85: based on the temperature value result of each pixel point, the image of the multi-phase component temperature result in the particle combustion process is sorted and displayed by utilizing a multi-phase component temperature result image integration algorithm in the particle combustion process.
The particle combustion process multi-phase component identification and image decomposition algorithm is characterized in that different phase component radiation images are identified according to a multi-band spectral response value clustering analysis algorithm of each pixel point of a combustion particle radiation image, and the images are decomposed into imaging backgrounds and particle combustion phase image areas.
The clustering analysis algorithm for the multiband spectral response values of all pixel points of the combustion particle radiation image is characterized in that the clustering analysis algorithm randomly selects the multiband spectral response values of all pixel points of the combustion particle radiation image from a data set containing the multiband spectral response valueskTaking the data samples as initial clustering centers, and counting each spectrum sample andkthe distance of each initial clustering center, all the spectral data are divided into the categories represented by the clustering centers with the closest distances to the initial clustering centers, and the spectral data are updated according to the mean value of the spectral samples in the newly generated categorieskAnd (4) clustering centers. If the change of the clustering center value in the adjacent iteration times exceeds the set threshold value, performing classification again on all the data samples according to the new clustering center; if the variation of the clustering central value in the adjacent iteration times is smaller than a specified threshold value, the algorithm converges and the clustering result is output, so that the identification of the multi-phase components in the particle combustion process is realizedRespectively with image decomposition.
The clustering analysis algorithm for multiband spectral response values of all pixel points of the combustion particle radiation image is characterized in that the processing flow of the algorithm is as follows:
(1) selecting a clustered original data set;
(2) from which random selection is madekUsing the data sample as initial clustering centerz 1, z 2, …, z k ；
(3) Calculating all data sample data one by one tokThe distance between each condensation point (usually Euclidean distance is used as the distance from the sample to the cluster center), and the distance is determined according to the size of the distancenEach sample (or variable) being divided intokClass, Euclidean distance calculation formula is as follows:
Figure 439526DEST_PATH_IMAGE011
wherein,x i is a samplexTo (1) aiThe value of a variable of the individual variables,y i is a sampleyIs/are as followsiThe variable values of the individual variables. If the distance from the data sample to the original class is the shortest, the data sample is still in the original class, otherwise, the data sample is moved to the class which is the shortest;
(4) computingkAnd (3) if the clustering center of each type of data in the class is not coincident with the initial clustering center, taking the clustering center as a new clustering center, repeating the step (3) until all the spectrum samples cannot move or each clustering center is not changed, and terminating the calculation process.
The radiation characteristic algorithm based on the radiation spectrum is characterized in that the radiation spectrum of each pixel point is processed by the algorithm, the Planck radiation law is adopted to describe the radiation spectrum of an object, the spectral radiation force is distributed along with the wavelength, and the spectral radiation force of a black body on each wave bandE b Can be expressed as:
Figure 389028DEST_PATH_IMAGE012
wherein,E b is the radiation intensity of black body in each wave band, W.m-2；C 1AndC 2respectively a planck first radiation constant and a second radiation constant,C 1=3.7419×10-16W·m-2、C 2=1.4388×10-2m·K；λis the wavelength, m;Tis the temperature, K;
the radiance of the material is constructed by adopting the following polynomial functionε(λ)：
Figure 150311DEST_PATH_IMAGE013
Wherein,a i is a constant of a polynomial and is,nis the fitting order;
and determining a wave band range meeting the radiation gray property according to a radiation gray property judgment method by obtaining the phase state radiance model to which each pixel point belongs.
The radiation grayness judgment method is characterized in that the ratio of monochromatic response intensities on two wavelengths is calculated:
Figure 476250DEST_PATH_IMAGE014
wherein,ε(λ)、ε(λ+Δλ) Respectively the burning particles at wavelengthλ、λ+ΔλThe radiance of (c). When the wavelength changes by an amount of ΔλSufficiently small, the temperature value can be calculated from the ratio of the radiation intensities of the two wavelengths:
Figure 639378DEST_PATH_IMAGE015
after the temperature value is obtained, the radiance distribution in the wavelength range can be calculated according to the response intensity of the measured object and the response intensity of the black body at the same temperature:
Figure 759781DEST_PATH_IMAGE016
wherein,I b (λ) Is the response value of the blackbody radiation spectrum;
according to the change condition of the radiance distribution along with the wavelength, whether the combustion particles meet the radiance of the ash body can be judged. If the emissivity changes more smoothly with wavelength, the soot body radiation characteristics of the combustion particles in this band are better, and if the change is more dramatic, they should not be considered soot bodies in this band. Analyzing the fluctuation of radiance distribution by using relative variance, and obtaining average radiance in a certain waveband rangeε(a)：
Figure 241315DEST_PATH_IMAGE017
Wherein,mis the wave number;
the variance of radiance over this band can be expressed as:
Figure 105366DEST_PATH_IMAGE018
with respect to the emissivity variance, it is generally considered that when it is less than 5%, the radiation characteristic of the combustion particles in this wavelength band satisfies the radiation characteristic of the soot body.
The radiation spectrum temperature inversion algorithm is characterized in that the algorithm obtains the temperature value of each pixel point based on a Planck's law parameter fitting algorithm according to the selected spectrum data in the wave band range meeting radiation grayness.
The Planck's law parameter fitting algorithm is characterized in that through obtaining photoelectric response values of all wave bands of all pixel points of a radiation image, response intensity can be searched by utilizing a least square method parameter fitting algorithmI(λ,T) AndE(λ,T) Optimal matching parameter betweenεAndT，the temperature is obtained.
The least square method parameter fitting algorithm is characterized by passing the minimumFunction model established by parameter fitting principle of two-timesF(ε, T) Analyzing the optimal solution of the temperature and emissivity parameters of the burning particles:
Figure 123001DEST_PATH_IMAGE019
in the formula,Iin order to obtain a spectral response value by measurement,kthe obtained response coefficient is calibrated through photoelectric response.
The particle combustion process multi-phase component temperature result image integration algorithm is characterized in that the algorithm integrates the multi-phase component temperature result of the particle combustion process according to the coordinates of all pixel points and the temperature measurement result to form a particle combustion process multi-phase component temperature distribution image.
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes performed by the present specification and drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.

Claims (10)

1. A multispectral reconstruction temperature measuring device for a particle combustion process is combined with direct imaging and spectral imaging technologies, direct radiation imaging and radiation multispectral imaging in the particle combustion process are synchronously registered, the correlation between a radiation image and the radiation multispectral image in the particle combustion process is trained and excavated by adopting an artificial neural network, and the multispectral reconstruction is carried out on the radiation image in the particle combustion process based on a multispectral reconstruction and temperature inversion algorithm, so that the online measurement of multicomponent characteristic temperature distribution in the particle combustion process is realized, and the multispectral reconstruction temperature measuring device is characterized by comprising: the particle combustion radiation imaging system comprises a quartz glass plate, a beam splitter prism, a spectral imaging unit, a direct imaging unit, a synchronous trigger, an image processing unit and a control and signal cable, wherein particle radiation light is split by the beam splitter prism through the quartz glass plate, one beam of light enters the spectral imaging unit to obtain a particle combustion radiation spectral image with high spectral resolution, and the other beam of light enters the direct imaging unit to obtain a particle combustion radiation image with high spatial resolution; the quartz glass plate is used for preventing combustion particles from polluting optical components; the synchronous trigger is used for synchronizing the simultaneous work of the spectral imaging unit and the direct imaging unit; the image processing unit is used for synchronizing direct radiation imaging and radiation multispectral imaging registration in the particle combustion process, adopting an artificial neural network to train and excavate the correlation between the radiation image and the radiation multispectral image in the particle combustion process, and carrying out multispectral reconstruction on the radiation image in the particle combustion process based on a multispectral reconstruction and temperature inversion algorithm so as to realize online measurement of multicomponent characteristic temperature distribution in the particle combustion process.
2. The multispectral reconstruction temperature measuring device for the particle combustion process as claimed in claim 1, wherein the spectral imaging unit comprises a spectral imaging attenuation sheet, a spectral imaging lens and a spectral camera, and after the particle radiation light is split by the beam splitter prism, a beam of light enters the spectral imaging attenuation sheet for attenuation, is adjusted by the spectral imaging lens and enters the spectral camera for imaging to obtain a multispectral image of the particle combustion process radiation; the direct imaging unit comprises a direct imaging attenuation sheet, a direct imaging lens and an industrial camera, wherein the particle radiation light is split by the light splitting prism, and then enters the direct imaging attenuation sheet for attenuation, is adjusted by the direct imaging lens, and enters the industrial camera for imaging to obtain a particle combustion process radiation image.
3. The device according to claim 1, wherein the image processing unit is connected to the spectral camera of the spectral imaging unit and the industrial camera of the direct imaging unit via signal cables, and is configured to perform on-line measurement of the multi-component characteristic temperature distribution of the particle combustion process by obtaining a synchronous image of direct radiation imaging and radiation multi-spectral imaging registration through the particle combustion process, training and mining the correlation between the radiation image of the particle combustion process and the radiation multi-spectral image using an artificial neural network, and performing multi-spectral reconstruction on the radiation image of the particle combustion process based on multi-spectral reconstruction and a temperature inversion algorithm.
4. The multispectral reconstruction temperature measurement device for the particle combustion process according to claim 1, wherein the spectral camera comprises a condenser lens, an optical slit, an adjusting lens, a grating, an imaging lens, a micro-array lens, a photoelectric detection array, a digital image acquisition unit and an image processor, wherein the condenser lens is used for adjusting a condensing imaging position plane of a particle radiation optical signal; the optical slits are used for reducing the influence of ambient light; the adjusting lens is used for adjusting the particle image radiation light to enter the grating in parallel; the grating is used for decomposing optical signals with different wavelengths; the imaging lens is used for adjusting the wavelength splitting signal to enter the micro-array lens; the micro-array lens is used for converging light to enter the photoelectric detection array; the photoelectric detection array is used for converting optical signals into electric signals; the digital image acquisition unit is used for converting the received photoelectric detection array electric signal into a digital image signal; the image processor is used for processing the received digital image signals to obtain particle combustion radiation spectrum images.
5. A particle combustion process multispectral reconstruction temperature measurement method is characterized by mainly comprising two processes of direct radiation imaging and radiation multispectral imaging registration synchronization in a particle combustion process, adopting artificial neural network training to excavate a particle combustion process radiation image and a radiation multispectral image, and carrying out multispectral reconstruction temperature measurement on the particle combustion process radiation image based on a multispectral reconstruction and temperature inversion algorithm, and specifically comprises the following steps:
s1: a particle combustion process multispectral reconstruction temperature measuring device is adopted, and a radiation image and a radiation multispectral image in the particle combustion process are synchronously obtained through direct radiation imaging and radiation multispectral imaging registration in the particle combustion process;
s2: normalizing the grain size of the particle combustion image;
s3: angle correction and translation correction of synchronous particle combustion images;
s4: target identification of particle combustion images;
s5: establishing an artificial neural network model of the spectrum dimension and the image dimension;
s6: training an artificial neural network model based on an error back propagation algorithm;
s7: optimizing the number of hidden layers and the number of neurons of the artificial neural network;
s8: and performing multispectral reconstruction on the combustion particle radiation image and inverting the temperature of the combustion particle radiation image.
6. The particle combustion process multispectral reconstruction thermometry method of claim 5, wherein the particle combustion process direct radiation imaging and radiation multispectral imaging registration synchronization process comprises:
the grain combustion image texture size normalization is to perform texture size normalization on response images of an industrial camera and a spectral camera sensor; the pixel resolution of the industrial camera image is higher than that of the spectral camera image, so the texture size normalization process is also a process of degrading the definition of the radiation image to the definition of the radiation multispectral image; the acquired radiation image can be represented in a matrix formX M N××3WhereinM×NFor its spatial resolution, the radiation multispectral image matrix is represented asY m n l××With a spatial resolution ofm×n，lIs the number of wave bands; need to be provided withXResolution ofM×NDegenerating tom×nThe pixels of the two camera image data sets can be subjected to model training, and the factors of the resolution ratio and the pixel area of the sensor and the degradation coefficient of the radiation image are considered in the processCExpressed as:
Figure 223171DEST_PATH_IMAGE001
wherein,S X is the pixel area of an industrial camera sensor,S Y is the pixel area of the spectral camera sensor, and the matrixX M N××3Degenerating toX m n××3The texture size normalization of the radiation image and the radiation multispectral image can be realized, so that each corresponding pixel point in the radiation image displays the same texture position of the combustion particles;
the angle correction and the translation correction of the synchronous particle combustion image are to solve the problem that the images of the same object obtained by the detection of two camera sensors may have angle deviation and translation amount after the particle combustion radiation signal passes through a beam splitting prism, a backlight method is adopted to shoot a calibration plate image, and the translation amount and the rotation angle between the two images are calculated by utilizing a cross-correlation template matching analysis method; the cross-correlation analysis of template matching is to define the radiation image and radiation multispectral image after graying as image template matrixXAndYmatching images by analyzing cross-correlation coefficients between templates; template matrixXThrough displacement and rotation change to obtainX i j θ,,The variation formula is as follows:
Figure 82543DEST_PATH_IMAGE002
wherein,m、nrepresenting a template matrixXThe column-row value of (a),iandjas a templateXThe amount of translation in the parallel and vertical directions,θis an anticlockwise rotating angle and obtains the template after conversionX i j θ,,；
ComputingX i j θ,,AndYcross correlation coefficient betweenR(i, j, θ) When the cross-correlation coefficient reaches a maximum value, corresponding toi, j, θ) Namely the templateXAndYthe amount of translation and the angle of rotation therebetween;
Figure 222537DEST_PATH_IMAGE003
therefore, the optimal solution of the translation amount and the angle is obtained through the cross-correlation analysis of the calibration plate image;
the particle combustion image target identification adopts a Canny edge processing method, the gradient amplitude and the direction of the gray level image are calculated, the point with the maximum local gradient is reserved by adopting a non-maximum inhibition method for the gradient value, and finally the image is thresholded.
7. The particle combustion process multispectral reconstruction thermometry method of claim 5, wherein said mining the particle combustion process radiation image and the radiation multispectral image using artificial neural network training comprises:
the establishing of the artificial neural network model of the spectrum dimension and the image dimension is to establish a mathematical relationship model between the direct radiation imaging and the radiation multispectral imaging based on the artificial neural network on the basis of the synchronization of the direct radiation imaging and the radiation multispectral imaging in the particle combustion process; in the process of multispectral reconstruction of the radiation image, response signals of the radiation image at each pixel point (x r , x g , x b ) And (3) as an input layer variable of the neural network, sending the variable into the model, wherein the relational expression of each neuron model in the first layer of the hidden layer is as follows:
Figure 163074DEST_PATH_IMAGE004
in the formula,x j response intensity on R, G, B wave band of each pixel point in the radiation image;w i j,to representx j The weight coefficient of (a);b i is composed ofx j A bias threshold of (a);kis composed ofb i Andw i j,the dimension of (a) is the number of neuron units in the hidden layer; the weight coefficients and bias thresholds of the neuron elements represent a linear processing of the data by the network,u i the sum of the two is obtained;
the artificial neural network model trained based on the error back propagation algorithm adopts a nonlinear function in the neural network model to utilize an excitation functionϕ(u) Expressing:
Figure 236072DEST_PATH_IMAGE005
will be provided withϕ(u) As the output value of the first layer of the hidden layer, the output value is sent to the next layer of the neural network, if the weight coefficient of the current layer is weightedw i j,And paranoia thresholdb i Is defined as a weight matrixWAnd a bias matrixBThe input value and the output value of the layer areXAndYa matrix, which can be expressed as:
Figure 735186DEST_PATH_IMAGE006
the specific matrix operation is as follows:
Figure 362477DEST_PATH_IMAGE007
the resulting matrixYSending to the next layer of the network model until reaching the output layer, and outputting the result (z 1, z 2, …, z m ) And realizing the feed-forward propagation of data in the artificial neural network, wherein,mis the dimension of the output layer;
calculating the error between the expected output value and the network output value by the artificial neural network model by using a loss function, and taking the radiation multispectral image corresponding to the radiation image as the expected output value of the neural network model (z’ 1, z’ 2, …, z’ m )；
The training results of the network were evaluated using Mean Squared Error (MSE) as a loss function:
Figure 605239DEST_PATH_IMAGE008
MSE is the error of the prediction result, and the weight coefficient of each layer in the network model is adjusted by using a gradient descent method according to the value of the errorWAnd bias thresholdBBy solving the partial derivatives of the mean square error to the parameters in the output layerNumber to determine the degree of adjustment of the parameters in the network model:
Figure 267165DEST_PATH_IMAGE009
Figure 937180DEST_PATH_IMAGE010
wherein,ηthe step length is adjusted, namely the learning rate of the neural network model is obtained;
the artificial neural network implicit layer number and neuron number optimization utilizes a gradient descent method to continuously adjust the weight and bias of the neuron unit from an output layer to an input layer, so that the MSE of the output value and the expected output value of the neural network is continuously reduced, reverse transmission of errors and iteration of an algorithm are realized until the MSE between the output value and the expected value of the network model reaches a preset target value, and training of the neural network is completed;
the multispectral reconstruction of the combustion particle radiation image is realized by adopting the radiation image after pixel degradation and texture information of combustion particles in the radiation multispectral radiation image to correspond one to one according to pixel points, and establishing a mathematical model through the mapping relation between the response intensity of the radiation image and the multispectral response intensity.
8. The method according to claim 5, wherein the multispectral reconstruction temperature measurement of the combustion particle radiation image comprises the following steps:
s81: acquiring a particle combustion radiation spectrum image by adopting a particle combustion process multispectral reconstruction temperature measuring device, and performing multispectral reconstruction;
s82: identifying the multiphase components in the particle combustion process and decomposing an image area by utilizing a multiphase component identification and image decomposition algorithm in the particle combustion process;
s83: processing the same-phase component image area in the particle combustion process, and obtaining the radiation characteristic of the phase component through the radiation spectrum of each pixel point by using a radiation characteristic algorithm based on the radiation spectrum;
s84: processing the radiation spectrum of each pixel point in the same-phase component image area in the particle combustion process, and obtaining a temperature value of the pixel point by using a radiation spectrum temperature inversion algorithm;
s85: based on the temperature value result of each pixel point, the image of the multi-phase component temperature result in the particle combustion process is sorted and displayed by utilizing a multi-phase component temperature result image integration algorithm in the particle combustion process.
9. The particle combustion process multi-spectral reconstruction temperature measurement method according to claim 5, wherein the particle combustion process multi-phase component identification and image decomposition algorithm identifies different phase component radiation images according to a multi-band spectral response value cluster analysis algorithm of each pixel point of a combustion particle radiation image, and decomposes the images into an imaging background and each particle combustion phase image region;
the clustering analysis algorithm for the multiband spectral response values of all pixel points of the combustion particle radiation image is characterized in that the clustering analysis algorithm randomly selects the multiband spectral response values of all pixel points of the combustion particle radiation image from a data set containing the multiband spectral response valueskTaking the data samples as initial clustering centers, and counting each spectrum sample andkthe distance of each initial clustering center, all the spectral data are divided into the categories represented by the clustering centers with the closest distances to the initial clustering centers, and the spectral data are updated according to the mean value of the spectral samples in the newly generated categorieskA cluster center; if the change of the clustering center value in the adjacent iteration times exceeds the set threshold value, performing classification again on all the data samples according to the new clustering center; if the variation of the clustering central value in the adjacent iteration times is smaller than a specified threshold value, the algorithm converges and a clustering result is output, so that the multi-phase component identification and the image decomposition in the particle combustion process are realized;
the clustering analysis algorithm for multiband spectral response values of all pixel points of the combustion particle radiation image is characterized in that the processing flow of the algorithm is as follows:
(1) selecting a clustered original data set;
(2) from which random selection is madekUsing the data sample as initial clustering centerz 1, z 2, …, z k ；
(3) Calculating all data sample data one by one tokThe distance between each condensation point (usually Euclidean distance is used as the distance from the sample to the cluster center), and the distance is determined according to the size of the distancenEach sample (or variable) being divided intokClass, Euclidean distance calculation formula is as follows:
Figure 284723DEST_PATH_IMAGE011
wherein,x i is a samplexTo (1) aiThe value of a variable of the individual variables,y i is a sampleyIs/are as followsiVariable values of the individual variables; if the distance from the data sample to the original class is the shortest, the data sample is still in the original class, otherwise, the data sample is moved to the class which is the shortest;
(4) computingkAnd (3) if the clustering center of each type of data in the class is not coincident with the initial clustering center, taking the clustering center as a new clustering center, repeating the step (3) until all the spectrum samples cannot move or each clustering center is not changed, and terminating the calculation process.
10. The particle combustion process multi-spectral reconstruction temperature measurement method according to claim 5, wherein the radiation characteristic algorithm based on radiation spectrum processes the radiation spectrum of each pixel point, and adopts Planck's radiation law to describe the radiation spectrum of the object, the spectral radiation power is distributed with the wavelength, and the spectral radiation power of the black body on each bandE b Can be expressed as:
Figure 331176DEST_PATH_IMAGE012
wherein,E b is the radiation intensity of black body in each wave band, W.m-2；C 1AndC 2respectively a planck first radiation constant and a second radiation constant,C 1=3.7419×10-16W·m-2、C 2=1.4388×10-2m·K；λis the wavelength, m;Tis the temperature, K;
the radiance of the material is constructed by adopting the following polynomial functionε(λ)：
Figure 909925DEST_PATH_IMAGE013
Wherein,a i is a constant of a polynomial and is,nis the fitting order;
determining a wave band range meeting radiation gray property according to a radiation gray property judgment method by obtaining a phase state radiance model to which each pixel point belongs;
the radiation grayness judgment method is characterized in that the ratio of monochromatic response intensities on two wavelengths is calculated:
Figure 750842DEST_PATH_IMAGE014
wherein,ε(λ)、ε(λ+Δλ) Respectively the burning particles at wavelengthλ、λ+ΔλThe emissivity of (d);
when the wavelength changes by an amount of ΔλSufficiently small, the temperature value can be calculated from the ratio of the radiation intensities of the two wavelengths:
Figure 87146DEST_PATH_IMAGE015
after the temperature value is obtained, the radiance distribution in the wavelength range can be calculated according to the response intensity of the measured object and the response intensity of the black body at the same temperature:
Figure 766651DEST_PATH_IMAGE016
wherein,I b (λ) Is the response value of the blackbody radiation spectrum;
according to the change condition of the radiance distribution along with the wavelength, whether the combustion particles meet the radiance of the ash body can be judged;
if the radiance changes more stably along with the wavelength, the radiation characteristic of the ash body of the combustion particles in the wave band is better, and if the radiance changes more severely, the combustion particles in the wave band are not regarded as the ash body;
analyzing the fluctuation of radiance distribution by using relative variance, and obtaining average radiance in a certain waveband rangeε(a)：
Figure 668748DEST_PATH_IMAGE017
Wherein,mis the wave number;
the variance of radiance over this band can be expressed as:
Figure 680566DEST_PATH_IMAGE018
regarding the radiance variance, it is generally considered that when it is less than 5%, the radiation characteristic of the combustion particle in this band satisfies the radiation characteristic of the soot body;
the radiation spectrum temperature inversion algorithm is characterized in that the algorithm obtains the temperature value of each pixel point based on a Planck's law parameter fitting algorithm according to the selected spectrum data in the wave band range meeting radiation grayness;
the Planck's law parameter fitting algorithm is characterized in that through obtaining photoelectric response values of all wave bands of all pixel points of a radiation image, response intensity can be searched by utilizing a least square method parameter fitting algorithmI(λ,T) AndE(λ,T) Optimal matching parameter betweenεAndT，the temperature can be obtained;
the least square method parameter fitting algorithm is characterized in that a function model is established by the least square parameter fitting principleF(ε, T) Analyzing the optimal solution of the temperature and emissivity parameters of the burning particles:
Figure 769745DEST_PATH_IMAGE019
in the formula,Iin order to obtain a spectral response value by measurement,kthe response coefficient obtained through photoelectric response calibration;
the particle combustion process multi-phase component temperature result image integration algorithm is characterized in that the algorithm integrates the multi-phase component temperature result of the particle combustion process according to the coordinates of all pixel points and the temperature measurement result to form a particle combustion process multi-phase component temperature distribution image.
CN202111474356.5A 2021-12-06 2021-12-06 Multispectral reconstruction temperature measuring device and method in particle combustion process Pending CN114061766A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111474356.5A CN114061766A (en) 2021-12-06 2021-12-06 Multispectral reconstruction temperature measuring device and method in particle combustion process

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111474356.5A CN114061766A (en) 2021-12-06 2021-12-06 Multispectral reconstruction temperature measuring device and method in particle combustion process

Publications (1)

Publication Number Publication Date
CN114061766A true CN114061766A (en) 2022-02-18

Family

ID=80228564

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111474356.5A Pending CN114061766A (en) 2021-12-06 2021-12-06 Multispectral reconstruction temperature measuring device and method in particle combustion process

Country Status (1)

Country Link
CN (1) CN114061766A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116543845A (en) * 2023-04-10 2023-08-04 中国科学院力学研究所 Visual analysis method and system for chemical reaction flow field
CN117313553A (en) * 2023-11-28 2023-12-29 四川物科光学精密机械有限公司 Multispectral radiation temperature measurement inversion calculation method based on neural network

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116543845A (en) * 2023-04-10 2023-08-04 中国科学院力学研究所 Visual analysis method and system for chemical reaction flow field
CN117313553A (en) * 2023-11-28 2023-12-29 四川物科光学精密机械有限公司 Multispectral radiation temperature measurement inversion calculation method based on neural network
CN117313553B (en) * 2023-11-28 2024-02-06 四川物科光学精密机械有限公司 Multispectral radiation temperature measurement inversion calculation method based on neural network

Similar Documents

Publication Publication Date Title
Manolakis et al. Longwave infrared hyperspectral imaging: Principles, progress, and challenges
US8050884B2 (en) Method and apparatus for determining the emissivity, area and temperature of an object
Jang et al. Experimental demonstration of adaptive infrared multispectral imaging using plasmonic filter array
CN114061766A (en) Multispectral reconstruction temperature measuring device and method in particle combustion process
Bernacki et al. Standoff hyperspectral imaging of explosives residues using broadly tunable external cavity quantum cascade laser illumination
AU2007217794A1 (en) Method for spectral data classification and detection in diverse lighting conditions
CN111795746B (en) Flame multi-parameter field cooperative measurement method based on active and passive optical chromatography fusion detection
Li et al. Rapid sensing of hidden objects and defects using a single-pixel diffractive terahertz sensor
CN115222707A (en) Full-spectrum hyperspectral earth remote sensing imaging measurement uncertainty analysis method
Breshike et al. Rapid detection of infrared backscatter for standoff detection of trace explosives
CN111289496A (en) Detection method and device for long-distance zoom laser-induced breakdown spectroscopy
Yousefi et al. Continuum removal for ground-based LWIR hyperspectral infrared imagery applying non-negative matrix factorization
EP0907883B1 (en) Method and system for remotely determining column density of trace gases
CN116295852A (en) Neutron scattering experimental sample temperature field monitoring method based on graph neural network
Kohler et al. New approach for the radiometric calibration of spectral imaging systems
Kozinov et al. Development and processing of hyperspectral images in optical–electronic remote sensing systems
CN113566971B (en) Multispectral high-temperature transient measurement system based on neural network
CN114689181A (en) Multispectral temperature measuring device, system and measuring method thereof
Lang et al. Classification of materials using a pulsed time-of-flight camera
Lapeyre et al. A Bayesian error model for quantifying methane and carbon dioxide column densities from hyperspectral measurements
Meng et al. Material identification by plasmonic infrared microspectrometer employing machine learning
Zhao et al. Application of hyperspectral imaging in measurement real-time of seeds
CN114485942B (en) Hyperspectral registration method and imaging system thereof
Kerekes et al. Modeling of LWIR hyperspectral system performance for surface object and effluent detection applications
CN113588115B (en) Temperature measurement method based on multispectral colorimetric

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination