Intelligent Analysis And Research On Food Safety Behavior Of Transparent Kitchen Based On Deep Learning

The people regard food as the sky and food as safety.For thousands of years,food safety has accompanied our food culture.Now,with the help of in-depth study and the increasingly perfect computer hardware equipment,we are improving and strengthening food supervision to protect food safety.This paper mainly aims at the "transparent kitchen" project,and adopts the deep learning method to realize the detection of the video collected by the kitchen and complete the identification of the nonstandard operation in the kitchen.In the early stage,the video is mainly collected and screened,then the video is preprocessed and redundant is removed,frames are extracted from the video as training images,renamed according to the format of VOC data sets,a large number of preprocessed images are labeled by Label Img,and exclusive data sets are made for model training.In order to achieve the purpose of real-time monitoring in the later application,on the Caffe framework,with the help of GPU-equipped servers,the environment is first built.After that,the original VGG 16 in Faster RCNN algorithm is replaced by Shuffle Net V2 to reduce convolution,thus greatly reducing convolution parameters,improving parallelism and taking into account recognition speed and accuracy.Based on the existing model,super parameters such as iteration number,batch size,learning rate,etc.are debugged,RPN network of Faster RCNN is trained,proposals are collected,Faster RCNN network is trained,and thus a mature model is obtained by repeating twice.Finally,the work clothes and work caps worn by the staff in the test data set were successfully identified.The identification speed was improved compared with the traditional Faster RCNN,and the accuracy rate of work clothes could reach more than90%,but the identification rate of caps was not satisfactory,and the accuracy of labeling boxes also needed to be improved.