Research And Implementation Of Optical Disc Classification System Based On Improved ResNet50

Image classification is one of the core tasks in the field of computer vision,and it is the basis of many visual tasks,such as image detection and image segmentation,which are inseparable from the image classification problem.The practical application of image classification is very wide,and the " Clear your plate " classification proposed in this paper is also an image classification based on deep learning algorithm.In the context of vigorously advocating “Clear your plate”campaign in the whole society,various catering service industries have posted slogans of “Clear your plate” campaign,but there are still problems such as lack of supervision,superficial application,lack of classification standards,and high cost of manual identification.After a comprehensive analysis of the above problems,this paper proposes a plate classification algorithm based on the improved Res Net50,and develops a plate classification system based on the We Chat applet.The main work and contributions of this paper are as follows:（1）Plate Datasets,a plate dataset for university restaurants,is produced.This paper collects pictures through self-developed image acquisition terminal,mobile phone photography and web crawler technology.Among them,the image acquisition terminal and mobile phone photography method are mainly used to collect plate pictures,while the web crawler technology method is mainly used to obtain environmental photos in life.（2）An improved Res Net50 algorithm with simplified attention mechanism is proposed.In this paper,the deep convolutional neural network model Res Net50 is used as the core algorithm for disc classification.After training,the accuracy of the model is about 96.875%.For simple three-class problems,there is still room for improvement in accuracy.Therefore,this paper adopts the method of adding attention mechanism to the network structure to enhance the ability of the network to extract key features,and further improve the classification accuracy.Then the use of the attention mechanism was improved.At present,the use of various attention mechanisms has the problem of repeated learning attention for repeated modules.This creates a misuse of attention and increases the burden on the network.For this reason,this paper proposes a simplified attention mechanism,using the learned attention for the repeated modules in the network,reducing the burden of the network learning attention,making the attention more concentrated,and improving the accuracy of the network again.This paper simplifies three attention mechanisms:channel attention,CBAM attention and coordinate attention,and the comparative experiments have verified the effectiveness of the improved Res Net50 algorithm with the simplified the attention mechanism.（3）The improved Res Net50 algorithm was applied to the “Clear your plate”campaign,and a plate classification system based on We Chat applet was developed.The system includes modules such as user management,image collection and uploading,plate classification,and points rankings.This system implements online real-time plate classification and records the number of times users have successfully clocked in.This is of great significance for improving users’ awareness of saving and practicing “Clear your plate” campaign,and provides important theoretical and technical support for the precise classification of dinner plates in “Clear your plate”campaign.