A Deep Learning Algorithm for Forest Image Classification

The code used the 50-layer Residual Network (ResNet50) as the feature extraction part of the forest image classification algorithm.

Since the forest fire data set has no public data set, the experiment obtained 175 forest images from the Internet as the experimental data set(150 training images, 25 test images). Among the 150 images in the forest image training set, there are 50 normal images, 50 smoke images, and 50 fire images, and their labels are set to 0, 1, and 2, respectively.

the overall accuracy reached 92%

step 1-> Download the pre-training model resnet50-19c8e357, and put it in the pretraing directory

Link：https://pan.baidu.com/s/164mSo-81ixu444tw0WXvRg

password：8j7o

step 2 -> python train.py

step 3 -> python test.py