| With the rapid development of China’s economy,people’s requirements for material conditions and living standards also increase,but there is an inevitable phenomenon of food waste.On April 29,2021,China officially implemented the anti food waste law,which means that strict economy and anti waste have become legal requirements.The anti food waste law focuses on solving the food waste at the consumer end.Therefore,using technical means to accurately and efficiently detect waste behavior is of great significance to combat waste.This paper combines waste behavior detection with deep learning.Aiming at the problems of large difference,multi category and overlap between samples in the process of waste behavior detection,the accurate identification of waste behavior is better realized by optimizing the model.The main contents of this paper are as follows。(1)Aiming at the detection effect caused by mutual occlusion of tableware in waste behavior detection in real scene,this paper proposes a food waste behavior detection model based on improved SSD.Firstly,aiming at the problem of insufficient feature extraction ability caused by insufficient depth of VGG16 network in SSD network,the residual network is used to replace the VGG16 backbone network in the original SSD model;Secondly,the weight of the parameters of the feature extraction network is screened by adding the SE module,so as to strengthen the important channel information in the training process of the model;Finally,the Bi FPN is used to fuse the bottom edge features and high-level semantic features of the six feature extraction layers in SSD network.The experimental results show that the m AP of the improved SSD target detection algorithm in the waste behavior data set is 88.49%,which is 5.09% higher than that of the original SSD.(2)According to the characteristics of waste behavior in complex scenes,this paper proposes a food waste behavior recognition model in complex scenes.Firstly,the efficiency of waste behavior detection is improved by using the advantages of lightweight backbone network and less parameters in YOLOv4-tiny target detection network;Secondly,the SPPF is used to optimize the backbone feature extraction network,enrich the extractable target features in the backbone network and enhance the scale invariance of the backbone network;Then,PSA is used to distinguish the importance of the parameters in the feature layer in space and channel direction,so as to further improve the expression ability of features;Finally,the EIOU loss function is used to improve the convergence speed of the model and the regression accuracy of the prediction frame.The experimental results show that the m AP of the improved YOLOv4-tiny target detection algorithm in the waste behavior data set in complex scenes is 93.35%,which is 4.58%higher than that of the original YOLOv4-tiny.(3)Based on the improved SSD and YOLOv4-tiny target detection algorithm,this paper uses python programming language for system development,and selects Py Qt visual development tool to build a waste behavior real-time detection system.The system can automatically identify the waste behavior in life,solve the detection difficulty of waste behavior to a certain extent,improve the detection efficiency,reduce the maintenance cost of the system,and has a certain practical application value. |
