The original training data and test data are stored in the datase file. We need to create this folder by ourselves and download the dataset on Baidu. Link: https://pan.baidu.com/s/1soaGBldT3QMwch0uv1pUw, extract code: aik7, which is used to generate TXT files. The training and test files under the dataset file are generated by merging the subfiles under the dataset column. The training data set and the test data set are stored separately, and the image_txt file is generated by calling the class function in the makedata.py file. Picture content is stored in tfrecords file
Forward.py: forward operating network (vgg), backward.py: reverse optimization, save the trained model to the model file
App.py tests the single picture under app file, handlevideo.py file processing video, and applies the trained model to video analysis. The accuracy of smoke recognition is 92%.