Design Of Binocular Vision Garbage Classification System Based On Deep Learning

In the next 25 years,the total amount of waste in less developed countries will increase dramatically.Our country is in a stage of rapid development,and urban garbage is increasing,and the phenomenon of garbage "invading cities" is becoming more and more serious.What my country needs most in terms of garbage collection is to use smart garbage collection technology to solve the problems of front-end garbage sorting and distributing and sorting collection mode.In recent years,deep learning technology and computer vision technology have matured.Image processing,computer vision,and target detection technology have been rapidly improved.Combining deep learning and binocular vision technology to identify and locate garbage,and efficiently solve the problem of garbage sorting and recycling,Is the development trend of garbage intelligent recycling.In view of the above background,this article focuses on the topic of " Garbage classification based on deep learning and binocular vision ".The research contents of this topic include:(1)For the garbage detection module,the garbage images collected by the three methods of this article are derived from the garbage data set of this article.First,an image preprocessing scheme is constructed to ensure the quality of the homemade garbage data set.The three pre-training methods of VGG19,Inception V3 and Resnet50 are compared and analyzed.Resnet50 is selected as the network feature extractor,and the most suitable Fine-tuning depth of the Resnet50 model is obtained using the migration learning strategy.Build a lightweight improved target detection network MSSD as the front-end basic network,compare and analyze the training effects of SE and CBAM,and embed the Res Net50+CBAM module in the network model to replace the original single Res Net50 module to enhance the acuity of the garbage detection module in processing images,And finally proved the effectiveness and superiority of the improved network architecture through experiments.(2)For the binocular positioning module,first complete the calibration and correction of the binocular camera,compare and study the characteristics of the three stereo matching algorithms of SGBM,BM and GC,and finally use the SGBM stereo matching algorithm to process related matching tasks,and propose the feature of adding orientation constraints Point matching and improved methods based on gray value ranging further strengthen the performance of the binocular positioning module.Experiments show that under the self-made garbage data set,combining the results of the garbage detection module and the improved binocular positioning algorithm can obtain the depth and distance information of the target garbage.(3)Integrate the garbage detection module and the binocular positioning module to construct a garbage identification and location system,and conduct the test and performance analysis of the garbage identification and location system in the context of a single target garbage and multiple target garbage.