CN107423815A - A kind of computer based low quality classification chart is as data cleaning method - Google Patents
A kind of computer based low quality classification chart is as data cleaning method Download PDFInfo
- Publication number
- CN107423815A CN107423815A CN201710665692.5A CN201710665692A CN107423815A CN 107423815 A CN107423815 A CN 107423815A CN 201710665692 A CN201710665692 A CN 201710665692A CN 107423815 A CN107423815 A CN 107423815A
- Authority
- CN
- China
- Prior art keywords
- sum
- data
- class
- convolutional neural
- neural networks
- 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.)
- Granted
Links
- 238000004140 cleaning Methods 0.000 title claims abstract description 35
- 238000000034 method Methods 0.000 title claims abstract description 16
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 52
- 238000012549 training Methods 0.000 claims abstract description 25
- 238000012360 testing method Methods 0.000 claims description 37
- 101100373202 Rattus norvegicus Cx3cl1 gene Proteins 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 4
- 230000009286 beneficial effect Effects 0.000 claims description 3
- 101100268668 Caenorhabditis elegans acc-2 gene Proteins 0.000 claims description 2
- 238000013459 approach Methods 0.000 claims description 2
- 238000002474 experimental method Methods 0.000 abstract description 4
- 230000008569 process Effects 0.000 abstract description 3
- 238000013528 artificial neural network Methods 0.000 description 7
- 230000006870 function Effects 0.000 description 7
- 230000007935 neutral effect Effects 0.000 description 5
- 238000011156 evaluation Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000005284 excitation Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 241000270322 Lepidosauria Species 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000013529 biological neural network Methods 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000012372 quality testing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of computer based low quality classification chart as data cleaning method, low quality classification view data from internet batch collection can effectively be cleaned, so as to obtain higher-quality view data, for training the higher disaggregated model of a discrimination.Detailed process includes：First directly train a preliminary convolutional neural networks using low quality classification view data, data are identified in itself with the network again, wash the low image or quantity to a certain extent of pseudo- probability that Model Identification is this class and be less than a certain degree of image category, repeat said process and reach default standard until obtaining the discrimination of all picture data types.Contrast experiment illustrates that the present invention can effectively lift the classification quality and identification level of view data.
Description
Technical field
A kind of cleaning method of the low quality classification view data based on convolutional neural networks, this method can be to from interconnection
The low quality classification view data of net batch collection is effectively cleaned, so as to obtain higher-quality view data, for instructing
Practice the higher disaggregated model of a discrimination, belong to artificial neural network technology field.
Background technology
Artificial neural network (Artificial Neural Network, i.e. ANN), it is artificial since being the 1980s
The study hotspot that smart field rises.It is abstracted from information processing angle to human brain neuroid, and it is simple to establish certain
Model, different networks is formed by different connected modes.Neutral net or class are also often directly referred to as in engineering and academia
Neutral net.Neutral net is a kind of operational model, is formed by being coupled to each other between substantial amounts of node (or neuron).Each
A kind of specific output function of node on behalf, referred to as excitation function (activation function).Company between each two node
Connect and all represent one for the weighted value by the connection signal, referred to as weight, this memory equivalent to artificial neural network.
The output of network then according to the connected mode of network, weighted value and excitation function it is different and different.And network itself is generally all
Certain algorithm of nature or function are approached, it is also possible to a kind of expression of logic strategy.
Convolutional neural networks (Convolutional Neural Network, CNN) are a kind of feedforward neural networks, and
One kind of artificial neural network, its artificial neuron can respond the surrounding cells in a part of coverage, for large-scale
Image procossing has outstanding performance.Because its unique network structure can be effectively reduced the complexity of Feedback Neural Network, mesh
Preceding convolutional neural networks have turned into the study hotspot of current speech analysis and field of image recognition.Its weights share network structure
It is allowed to be more closely similar to biological neural network, reduces the complexity of network model, reduces the quantity of weights.The advantage is in network
Becoming apparent of showing when being multidimensional image of input, allow input of the image directly as network, avoid tional identification
Complicated feature extraction and data reconstruction processes in algorithm.Convolutional neural networks are compared to traditional neutral net, and its feature is such as
Under：
1. partially connected (Sparse Connectivity)
Convolutional network utilizes the space of image local special by being forced between adjacent two layers using local connection mode
Property, only it is connected in the Hidden unit of m layers with the regional area of the input block of m-1 layers, these partial zones of m-1 layers
Domain is referred to as the continuous acceptance region in space.
2. weights share (Shared Weights)
In convolutional neural networks, each sparse filter can cover whole visible range by shared weights, and these are common
The unit for enjoying weights forms a Feature Mapping, along with the cooperation with partially connected, constitutes feature extraction layer --- convolution
Layer.
3. pond layer (Pooling Layer)
Pond layer is another structure block of convolutional neural networks, and its function is the space chi by progressively reducing sign
The very little calculating to reduce in parameter amount and network.Pond layer independent operation on each characteristic pattern.
In addition, convolutional neural networks also include the element of traditional neutral net, such as full articulamentum and common
Nonlinear activation function sigmoid, tanh, ReLU etc..
Today of immense success is obtained in convolutional neural networks, a good data set is the convolutional Neural net trained
The key point of network model.Common data set has PASCAL VOC, MNIST, ImageNet, CIFAR-10 etc., wherein
ImageNet has common 15M high-resolution tape label images of 22K kinds, and image is collected in network, handmarking, is commonly used for entirely
The classification performance detection of convolutional neural networks model.
Data described above are general and professional, have passed through substantial amounts of inspection and handmarking.But for common
The data of application level, certain the class image that can be got are probably derived from internet reptile, are necessarily mingled among these a lot
Noise, the higher data of mass how are cleaned therefrom and give certain evaluation method, be the emphasis place of the present invention.From
After the higher data of quality are obtained in noise data, then the training of convolutional neural networks can be used it for, so as to reach one
The purpose of fixed practical application.
The content of the invention
The technical solution adopted by the present invention is a kind of cleaning of the low quality classification view data based on convolutional neural networks
Method, bag
Include following steps：
A) from internet, batch downloads the view data with label, and arrangement obtains the image data set of M classes altogether
DataSet0, wherein the picture number that the i-th class includes is Ni, i=1,2,3 ... M；
B) a convolutional neural networks CNN0 is trained with DataSet0, comprised the following steps that：
I. a convolutional neural networks model is built, and the structure for fixing this network model keeps constant；
Ii. training set of the certain proportion (such as 80%, 90%) as convolutional neural networks is taken at random to DataSet0；
Iii. the test set using the part of non-training set in DataSet0 as convolutional neural networks；
Iv. CNN0 is trained, iterates to after predetermined number of times and network test discrimination is designated as Acc0；
C) it is N to the i-th class image configuration length in DataSet0iOne dimensional image self-identifying array Ki, specific steps
It is as follows：
I. DataSet0 view data is identified with CNN0, is kth class wherein the i-th class jth image recognition
Pseudo- probability be designated as pijk, k=1,2,3 ... M, and these pseudo- probability are sorted from big to small；
If k=i, note self-identifying rate K in preceding L (such as L=10) pseudo- probability after ii. sorting be presentij=pijk, otherwise
Remember Kij=0；
D) self-identifying array K is analyzed, the low quality part cleaned in the i-th class view data：
I. the average value of the i-th class image self-identifying rate is calculated：
Ii. the standard deviation of the i-th class image self-identifying rate is calculated：
Iii. the cut off value SepVal=μ-σ * α of the i-th class image " low discrimination " are calculated, 1≤α≤10 and be integer, and
SepVal>0；
Iv. in the i-th class image, if there is Kij<SepVal, then wash jth image；Data set is obtained after the completion of cleaning
DataSet1；
E) convolutional neural networks for being carried out the same manner again using DataSet1 are trained, and obtain network test discrimination
Acc1, record simultaneously relatively and confirm whether cleaning is effective with Acc0；
F) in DataSet1, the i-th class amount of images is counted again, remembers that per class amount of images be N'i, to N'iEnter
Row is analyzed and cleans a small number of classifications, to reduce influence of the low quality data class to convolutional neural networks：
I. the average value of current M classifications amount of images is calculated：
Ii. the standard deviation of current M classifications amount of images is calculated：
Iii. the cut off value SepVal=μ-σ * α of " minority class " amount of images are calculated, 1≤α≤10 and be integer, and
SepVal>0；
Iv. the common m classes of classification that categorical measure in M class images is less than SepVal are counted；
V. remember that m class quantity summation is sum, M class quantity summation is SUM；
If vi. m/M is much larger than sum/SUM, judge the m classes for minority class, it is necessary to which cleaning treatment is fallen；
If m/M and sum/SUM numerical value approaches, then it is assumed that m class quantity is normal, without cleaning treatment.
G) convolutional neural networks for carrying out the same manner again with the data set DataSet2 after cleaning are trained, and obtain net
Network tests discrimination Acc2, and record simultaneously relatively and confirms whether cleaning is effective with Acc1；
H) according to the data obtained collection situation, repeat step (d) and (f), the common m ' classes of data category after being cleaned, m '<
M；
I) evaluated for the quality of the remaining common sum ' of m ' classes after cleaning view data：
I. all data of the m ' classes in DataSet0 are obtained, note total quantity is SUM ', SUM '>sum′；
Ii. the convolutional neural networks for the same manner being carried out to the m ' classes view data that total amount is SUM ' and sum ' are trained, and are obtained
To network test discrimination Acc (SUM ') and Acc (sum '), if Acc (SUM ')<Acc (sum '), the then data after explanation is cleaned
It is more beneficial for the classification based training of convolutional neural networks；
Iii. certain data test is randomly or manually extracted as common test from the m ' class data that total amount is sum '
Collection, using the data of SUM ' and the middle removing test parts of sum ' as training set, the convolutional neural networks training of the same manner is carried out,
It is Acc (SUM ') and Acc (sum ') to obtain network test discrimination；If Acc (SUM ')<Acc (sum '), then explanation is for same
Test set, using the data through over cleaning as training set train obtained by convolutional neural networks generalization ability it is stronger, survey
Examination discrimination is higher, i.e., the quality of data is higher.
Brief description of the drawings
Fig. 1 is experiment Integral Thought flow chart.
Fig. 2 is initial data set situation and its convolutional neural networks test discrimination result figure.
Fig. 3 is current data set self-identifying array schematic diagram.
Fig. 4 is to wash the data set situation and its convolutional neural networks test discrimination result after low-quality image
Figure.
Fig. 5 is analysis and the classification cleaning situation result figure that for the first time data are carried out with minority class.
Fig. 6 is the data set situation and its convolutional neural networks test discrimination result once cleaned after a small number of classifications
Figure.
Fig. 7 is that second of analysis that data are carried out with minority class and classification clean situation result figure.
Fig. 8 is the data set situation and its convolutional neural networks test discrimination result after secondary cleaning minority classification
Figure.
Fig. 9 is the comparing result figure for carrying out quality testing to cleaning front and rear class respectively.
Figure 10 is to use same test set, and data cleansing is front and rear to be used as training set, and the convolutional neural networks of training compare
Result figure.
Embodiment
Below in conjunction with the accompanying drawings and specific implementation case the invention will be further described：
1. downloading plants and flowers view data in batches from internet, arrangement can obtain 775 classification and amount to 161015 figures,
The picture number that wherein the i-th class includes is Ni(i=1,2,3 ... M)；
2. using gained image data set to train a convolutional neural networks, comprise the following steps that：
A) the network model file of AlexNet on python caffe is obtained, and obtains its pre- instruction on ImageNet
Practice model file, the initialization for convolutional neural networks；
B) view data concentrate it is random take about 90% data totally 144921 figures be used as training set, remaining 10%
16067 figures are used as test set, carry out convolutional neural networks training using caffe, after iteration 10000 times, obtain network test
Discrimination is 39%；
C) it is N that data are concentrated with the i-th class image configuration lengthiOne dimensional image self-identifying array Ki(i=1,2,3 ... M),
Tool
Body step is as follows：
I. original data collection is identified one by one with the convolutional neural networks that this is trained；
Ii. following processing is done to the pseudo- probability recognition result of the jth figure of the i-th class：
If 1) there is no the i-th class in the first 10 pseudo- probability recognition results that convolutional neural networks return, K is rememberedij=0；
2) if it is p to have the i-th class and probability in the first 10 pseudo- probability recognition results returned, K is rememberedij=p；
D) self-identifying array is analyzed, low-quality image data cleansing is carried out to i classes data：
I. the average value of the i-th class image self-identifying rate is calculated：
Ii. the standard deviation of the i-th class image self-identifying rate is calculated：
Iii. the cut off value SepVal=μ-σ * α (this experiment takes α=1) of " minority class " amount of images are calculated；
Iv. by the i-th class image, image of the discrimination less than SepVal is left out as low-quality image.
E) cleaning remaining 79198 figures later, about 90% is taken 71298 be used as training set, remaining 7900 as survey
Examination collection, carries out convolutional neural networks training with same method again, and it is 59.7% to obtain testing discrimination, compared to initial
39% has not small raising；
F) quantity of this 775 class image after cleaning is counted again, is designated as N'i, to N'iRow is analyzed and cleans a small number of classifications,
To reduce influence of the low quality data class to sorter network：
I. the average value of current M=775 classifications amount of images is calculated：
Ii. the standard deviation of current M=775 classifications amount of images is calculated：
Iii. the cut off value SepVal=μ-σ * α (this experiment takes α=1) of " minority class " amount of images are calculated；
Iv. after the cleaning of low quality noise image, statistics less than SepVal images classification number totally 178 classes, altogether
1815 figures；Because 178/775 is much larger than 1815/79198, therefore judges that these classes for a small number of low quality data classes, are cleaned
Fall；
G) 77383 view data after cleaning take 70000 to be used as training set, and remaining 7383 are used as test set,
Convolutional neural networks training is carried out with same method again, it is 60.0% to obtain testing discrimination, than the test of network before
Discrimination is slightly lifted；
H) according to the data obtained collection situation, minority class is cleaned again, and step is same as above, and the data category after being cleaned is total to
468 classes；
I) it is as follows for the performance evaluation of totally 70755 view data of remaining 468 classes after cleaning：
I. all data of 468 classes in initial data are obtained, total quantity is 111290；
Ii. the convolutional neural networks for 468 class data that total amount is 111290 and 70755 each being carried out with same method are instructed
Practice, it is 60.8% and 62.6% to obtain network test discrimination, illustrates that the data after cleaning are more beneficial for convolutional neural networks
Classification based training；
Iii. 10% is randomly selected as public test set test from the 468 class data that total amount is 70755, with
Test data are removed in 111290 and 70755 as training set, with same method training convolutional neural networks, obtained net
Network test discrimination is respectively 59.6% and 62.6%.This explanation, for same test set, the data network through over cleaning is general
Change ability is stronger, and Average Accuracy is higher, i.e., data performance is more preferable.
By experimental result it can be seen that：
1. this data cleansing effect is authentic and valid, by same evaluation method compared with having been lifted on Raw data quality.
2. next step cleaning strategy can determine according to current data set situation, this method is more flexible.
3. in the case of test set identical, the discrimination that the data after cleaning obtain convolutional neural networks is higher, explanation
The quality of data increases after cleaning.
Above example not limits technical scheme described in the invention only to illustrate the present invention.Therefore, all are not
Depart from technical scheme and its improvement of the spirit and scope of the present invention, all should cover among scope of the presently claimed invention.
Claims (1)
1. a kind of computer based low quality classification chart is as data cleaning method, it is characterised in that：This method includes following step
Suddenly, a) from internet, batch downloads the view data with label, and arrangement obtains the image data set DataSet0 of M classes altogether,
The picture number that wherein the i-th class includes is Ni, i=1,2,3 ... M；
B) a convolutional neural networks CNN0 is trained with DataSet0, comprised the following steps that：
I. a convolutional neural networks model is built, and the structure for fixing this network model keeps constant；
Ii. training set of the certain proportion as convolutional neural networks is taken at random to DataSet0；
Iii. the test set using the part of non-training set in DataSet0 as convolutional neural networks；
Iv. CNN0 is trained, iterates to after predetermined number of times and network test discrimination is designated as Acc0；
C) it is N to the i-th class image configuration length in DataSet0iOne dimensional image self-identifying array Ki, comprise the following steps that：
I. DataSet0 view data is identified with CNN0, the puppet that wherein the i-th class jth image recognition is kth class
Probability is designated as pijk, k=1,2,3 ... M, and these pseudo- probability are sorted from big to small；
If k=i, note self-identifying rate K in preceding L pseudo- probability after ii. sorting be presentij=pijk, otherwise remember Kij=0；
D) self-identifying array K is analyzed, the low quality part cleaned in the i-th class view data：
I. the average value of the i-th class image self-identifying rate is calculated：
Ii. the standard deviation of the i-th class image self-identifying rate is calculated：
Iii. the cut off value SepVal=μ-σ * α of the i-th class image " low discrimination " are calculated, 1≤α≤10 and be integer, and
SepVal>0；
Iv. in the i-th class image, if there is Kij<SepVal, then wash jth image；Data set is obtained after the completion of cleaning
DataSet1；
E) convolutional neural networks for being carried out the same manner again using DataSet1 are trained, and obtain network test discrimination Acc1,
Record simultaneously relatively and confirms whether cleaning is effective with Acc0；
F) in DataSet1, the i-th class amount of images is counted again, remembers that per class amount of images be N'i, to N'iDivided
Analyse and clean a small number of classifications, to reduce influence of the low quality data class to convolutional neural networks：
I. the average value of current M classifications amount of images is calculated：
Ii. the standard deviation of current M classifications amount of images is calculated：
Iii. the cut off value SepVal=μ-σ * α of " minority class " amount of images are calculated, 1≤α≤10 and be integer, and SepVal>
0；
Iv. the common m classes of classification that categorical measure in M class images is less than SepVal are counted；
V. remember that m class quantity summation is sum, M class quantity summation is SUM；
If vi. m/M is much larger than sum/SUM, judge the m classes for minority class, it is necessary to which cleaning treatment is fallen；
If m/M and sum/SUM numerical value approaches, then it is assumed that m class quantity is normal, without cleaning treatment；
G) convolutional neural networks for carrying out the same manner again with the data set DataSet2 after cleaning are trained, and obtain network survey
Discrimination Acc2 is tried, record simultaneously relatively and confirms whether cleaning is effective with Acc1；
H) according to the data obtained collection situation, repeat step (d) and (f), the common m ' classes of data category after being cleaned, m '<M；i)
Evaluated for the quality of the remaining common sum ' of m ' classes after cleaning view data：
I. all data of the m ' classes in DataSet0 are obtained, note total quantity is SUM ', SUM '>sum′；
Ii. the convolutional neural networks for the same manner being carried out to the m ' classes view data that total amount is SUM ' and sum ' are trained, and obtain net
Network test discrimination Acc (SUM ') and Acc (sum '), if Acc (SUM ')<Acc (sum '), the then data after explanation is cleaned more have
Beneficial to the classification based training of convolutional neural networks；
Iii. certain data test is randomly or manually extracted as common test collection from the m ' class data that total amount is sum ', with
The data of SUM ' and the middle removing test parts of sum ' carry out the convolutional neural networks training of the same manner, obtained as training set
Network test discrimination is Acc (SUM ') and Acc (sum ')；If Acc (SUM ')<Acc (sum '), then explanation is for same survey
Examination collection, the convolutional neural networks generalization ability obtained by being trained using the data through over cleaning as training set is stronger, and test is known
Not rate is higher, i.e., the quality of data is higher.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710665692.5A CN107423815B (en) | 2017-08-07 | 2017-08-07 | Low-quality classified image data cleaning method based on computer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710665692.5A CN107423815B (en) | 2017-08-07 | 2017-08-07 | Low-quality classified image data cleaning method based on computer |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107423815A true CN107423815A (en) | 2017-12-01 |
| CN107423815B CN107423815B (en) | 2020-07-31 |
Family
ID=60436570
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710665692.5A Expired - Fee Related CN107423815B (en) | 2017-08-07 | 2017-08-07 | Low-quality classified image data cleaning method based on computer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107423815B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108052925A (en) * | 2017-12-28 | 2018-05-18 | 江西高创保安服务技术有限公司 | A kind of cell personnel archives intelligent management |
| CN108055529A (en) * | 2017-12-25 | 2018-05-18 | 国家电网公司 | Electric power unmanned plane and robot graphics' data normalization artificial intelligence analysis's system |
| CN108334895A (en) * | 2017-12-29 | 2018-07-27 | 腾讯科技（深圳）有限公司 | Sorting technique, device, storage medium and the electronic device of target data |
| CN108596338A (en) * | 2018-05-09 | 2018-09-28 | 四川斐讯信息技术有限公司 | A kind of acquisition methods and its system of neural metwork training collection |
| CN108830294A (en) * | 2018-05-09 | 2018-11-16 | 四川斐讯信息技术有限公司 | A kind of augmentation method of image data |
| CN109190666A (en) * | 2018-07-30 | 2019-01-11 | 北京信息科技大学 | Flowers image classification method based on improved deep neural network |
| RU2732895C1 (en) * | 2019-05-27 | 2020-09-24 | Общество с ограниченной ответственностью "ПЛАТФОРМА ТРЕТЬЕ МНЕНИЕ" | Method for isolating and classifying blood cell types using deep convolution neural networks |
| CN113033694A (en) * | 2021-04-09 | 2021-06-25 | 深圳亿嘉和科技研发有限公司 | Data cleaning method based on deep learning |
| US11138478B2 (en) | 2018-06-08 | 2021-10-05 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method and apparatus for training, classification model, mobile terminal, and readable storage medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106649610A (en) * | 2016-11-29 | 2017-05-10 | 北京智能管家科技有限公司 | Image labeling method and apparatus |
-
2017
- 2017-08-07 CN CN201710665692.5A patent/CN107423815B/en not_active Expired - Fee Related
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106649610A (en) * | 2016-11-29 | 2017-05-10 | 北京智能管家科技有限公司 | Image labeling method and apparatus |
Non-Patent Citations (4)
| Title |
|---|
| TONG XIAO: "Learning from massive noisy labeled data for image classification", 《IEEE》 * |
| 李明: "数据清洗技术在文本挖掘中的应用", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 米允龙: "海量数据挖掘过程相关技术研究进展", 《计算机科学与探索》 * |
| 许杰: "基于医疗数据挖掘的在线病情分析系统研究与开发", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108055529A (en) * | 2017-12-25 | 2018-05-18 | 国家电网公司 | Electric power unmanned plane and robot graphics' data normalization artificial intelligence analysis's system |
| CN108052925A (en) * | 2017-12-28 | 2018-05-18 | 江西高创保安服务技术有限公司 | A kind of cell personnel archives intelligent management |
| CN108052925B (en) * | 2017-12-28 | 2021-08-03 | 江西高创保安服务技术有限公司 | Intelligent management method for community personnel files |
| CN108334895A (en) * | 2017-12-29 | 2018-07-27 | 腾讯科技（深圳）有限公司 | Sorting technique, device, storage medium and the electronic device of target data |
| CN108596338A (en) * | 2018-05-09 | 2018-09-28 | 四川斐讯信息技术有限公司 | A kind of acquisition methods and its system of neural metwork training collection |
| CN108830294A (en) * | 2018-05-09 | 2018-11-16 | 四川斐讯信息技术有限公司 | A kind of augmentation method of image data |
| US11138478B2 (en) | 2018-06-08 | 2021-10-05 | Guangdong Oppo Mobile Telecommunications Corp., Ltd. | Method and apparatus for training, classification model, mobile terminal, and readable storage medium |
| CN109190666A (en) * | 2018-07-30 | 2019-01-11 | 北京信息科技大学 | Flowers image classification method based on improved deep neural network |
| CN109190666B (en) * | 2018-07-30 | 2022-04-29 | 北京信息科技大学 | Flower image classification method based on improved deep neural network |
| RU2732895C1 (en) * | 2019-05-27 | 2020-09-24 | Общество с ограниченной ответственностью "ПЛАТФОРМА ТРЕТЬЕ МНЕНИЕ" | Method for isolating and classifying blood cell types using deep convolution neural networks |
| WO2020242341A1 (en) * | 2019-05-27 | 2020-12-03 | Общество С Ограниченной Ответственностью "Лаб Кмд" | Method for selecting and classifying blood cell types by means of deep convolutional neural networks |
| CN113033694A (en) * | 2021-04-09 | 2021-06-25 | 深圳亿嘉和科技研发有限公司 | Data cleaning method based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107423815B (en) | 2020-07-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107423815A (en) | A kind of computer based low quality classification chart is as data cleaning method | |
| CN105184312B (en) | A kind of character detecting method and device based on deep learning | |
| Wang et al. | Classification of white blood cells with patternnet-fused ensemble of convolutional neural networks (pecnn) | |
| CN105701480B (en) | A kind of Video Semantic Analysis method | |
| CN106407986A (en) | Synthetic aperture radar image target identification method based on depth model | |
| CN107871316B (en) | Automatic X-ray film hand bone interest area extraction method based on deep neural network | |
| CN109522924A (en) | A kind of broad-leaf forest wood recognition method based on single photo | |
| CN111340133A (en) | Image classification processing method based on deep convolutional neural network | |
| Nurhadiyatna | Deep learning for handwritten Javanese character recognition | |
| CN110490227A (en) | A kind of few sample image classification method based on Feature Conversion | |
| CN113295702B (en) | Electrical equipment fault diagnosis model training method and electrical equipment fault diagnosis method | |
| CN111582396A (en) | Fault diagnosis method based on improved convolutional neural network | |
| CN110455512B (en) | Rotary mechanical multi-integration fault diagnosis method based on depth self-encoder DAE | |
| CN111291727A (en) | Method and device for detecting signal quality by photoplethysmography | |
| CN115545086B (en) | Migratable feature automatic selection acoustic diagnosis method and system | |
| CN110327033B (en) | Myocardial infarction electrocardiogram screening method based on deep neural network | |
| CN107742132A (en) | Potato detection method of surface flaw based on convolutional neural networks | |
| CN107220707A (en) | Dynamic neural network model training method and device based on 2-D data | |
| Banerjee et al. | Enhancing Snake Plant Disease Classification through CNN-Random Forest Integration | |
| Ma et al. | Sprouting potato recognition based on deep neural network GoogLeNet | |
| CN116188834B (en) | Full-slice image classification method and device based on self-adaptive training model | |
| Li et al. | Fine-grained classification of grape leaves via a pyramid residual convolution neural network | |
| CN114118157A (en) | Illumination information diagnosis method and system based on plant electric signals | |
| CN112818982A (en) | Agricultural pest image detection method based on depth feature autocorrelation activation | |
| CN102855479A (en) | Printed Chinese character recognition system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20200731 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |