Research On Garbage Detection Method Based On Deep Convolutional Neural Networks

With the rapid development of economy and society,people’s material needs are increasing day by day,garbage piled up into mountains and improper disposal have caused great damage to the environment.Correctly identifying garbage and reducing the amount of garbage handled are conducive to resource conservation and recycling,and also protects the ecological environment,while improving the overall quality of the people and the level of social development.However,existing garbage detection technologies generally have problems such as insufficient accuracy and failing to meet real-time requirements.In order to solve the above problems and avoid waste of resources,this paper explores garbage detection based on the study of deep convolutional neural networks and object detection algorithms,mainly focusing on the garbage video images on the conveyor belt,and constructs three detection methods.The main research work is as follows:1)Garbage detection based on the YOLO model.This paper researches a garbage detection model based on YOLOv3.Dark Net-53 is used as the backbone network of YOLOv3,and the residual module is used as the feature extraction module to predict in a multi-scale manner.Experiments have proved that the network can effectively identify the location and category of the target in the self-made garbage data set,and shows good performance in garbage detection,but it still has the problem of not being able to effectively learn key features and identifying the location of objects.2)Garbage detection optimized based on YOLO Backbone.In this paper,VGGNet and Dense Net are used as the backbone network of YOLOv3 and applied to the garbage data set.As a basic feature extraction model,VGGNet has a simple network structure.Experimental results show that the algorithm has fast detection speed and can easily process large batches of image data.However,the detection accuracy of the VGGNet network is not enough,and sometimes there are missed detections.Dense Net,like Dark Net-53,draws on the residual idea of Res Net,but Dense Net benefits from the design of Dense Block,which improves the ability to extract shallow features and makes the connection between network levels closer.Experimental datas show that the algorithm improves the detection accuracy to a certain extent,but the Dense Net network structure is too complicated,and it is too expensive to use a lot of calculation and memory in exchange for a little accuracy.3)Garbage detection based on YOLO-Attention.In response to the above problems,this paper combines the attention mechanism and embeds the CBAM(Convolution Block Attention Module)module into the YOLOv3 network feature extraction module to focus on the key areas and channels in the image,and researches a YOLO-Attention garbage detection model.In order to verify the effectiveness of the researched method,this paper conducts experimental verification on the garbage data set.Comparing and analyzing the YOLO-Attention algorithm model with the "1)2)" algorithm,it proves that the YOLO-Attention algorithm researched in this paper has a better detection effect in the case of garbage data.