Design And Implementation Of The Garbage Classification System Based On Deep Learning

The industrial revolution makes the level of human productivity increase exponentially,and also makes the quantity of garbage increase rapidly.How to deal with garbage has become a thorny problem all over the world.As one of the important links of resource recovery and utilization,garbage classification can effectively improve the efficiency of resource recovery and utilization,and reduce the harm of environmental pollution.Traditional image classification algorithm is difficult to meet the requirements of garbage sorting equipment.With the development of deep learning technology,it is possible to automatically sort garbage with the help of visual technology.Taking pictures of garbage by camera,using convolution neural network to detect the category and location of garbage,and using manipulator or push plate to automatically complete the sorting task,can reduce the labor cost and improve the sorting efficiency.Therefore,the research of garbage image classification algorithm has theoretical significance and important application value.This paper combines the target detection algorithm and video tracking technology,simulates the scene of garbage classification in the environment of conveyor belt,realizes the automatic positioning of garbage,and provides the coordinates of garbage to the manipulator or mechanical pusher behind to complete the sorting process.The main contents of this paper include:(1)According to Huawei's garbage classification open data set,the garbage detection data set is marked and established;(2)Finally,resnet101 is selected as the main network of detection,and the technical scheme of adding attention mechanism and feature fusion mechanism into the network is proposed,which can better extract the garbage image information and complete the ablation experiment;(3)SSD is selected as the baseline of detection network,and model compressiontechnology is used to improve the real-time identification of garbage category and location;(4)This paper studies the methods of multi-target video tracking-sort method and deep sort method,analyzes the shortcomings of these two methods,and uses the improved deep sort algorithm to complete the video tracking of garbage on the conveyor belt,and achieves better accuracy and real-time performance;(5)The overall design of the garbage classification system is completed,and the classification module is realized.Through field experiments,the effectiveness of the algorithm is verified,which can basically meet the actual needs.