Research And Application Of Detection Algorithms For Intelligent Garbage Sorting

The rapid development of the social economy has greatly improved the living standards of residents,but at the same time,it has brought more and more household waste,and waste treatment has become increasingly important.The mid range waste sorting process of the waste treatment system relies on a large amount of manpower,resulting in low sorting efficiency,high false detection rate,and an unsuitable working environment for long-term workers.Therefore,achieving automatic waste sorting has become the key to promoting efficient waste sorting.To achieve automatic garbage sorting,this article proposes a deep learning based garbage detection algorithm,and based on this algorithm,simulates the actual garbage sorting pipeline,designs and builds an experimental automatic garbage sorting system.The main tasks are as follows:(1)In view of the problem that there is no public garbage detection data set at home and abroad at present,this paper marks the Huawei garbage classification data set manually,and uses data expansion and CutMix to simulate garbage occlusion to improve the sample generalization ability by increasing the number of samples,so as to obtain a higher quality self-made garbage detection data set.On this basis,this paper uses the dichotomous K-means algorithm which is better than the original clustering algorithm to re cluster the self-made garbage detection data set to obtain an anchor frame more suitable for this data set.(2)The algorithm in this article needs to be ported to embedded devices with limited computing power,ensuring detection speed and balancing detection accuracy.Therefore,the lightweight network GhostNet is used to replace the backbone of the baseline network YOLOv4 to improve detection speed;To address the issue of reduced detection accuracy caused by lightweight,a new M-ghost bottleneck structure is designed using MobileNeXt,and a large residual edge structure is added to the back three layers of the backbone network,with hierarchical adjustments to improve detection accuracy;In response to the problems of multiscale,large intra class differences,and mutual occlusion of garbage objects that affect detection accuracy,this paper introduces an improved adaptive feature fusion network to improve the detection ability of garbage at different scales.The channel global attention mechanism improves the network’s feature extraction ability and improves the problem of large intra class differences.Based on Soft-NMS and Diou-NMS,an E-NMS algorithm is proposed to improve the mutual occlusion problem between garbage;The experimental results show that the text network has a better performance than the baseline network mAP@0.5 The value increased by 1.9%,the parameter count decreased by 87.1%,and the FPS increased from 44.5 to 65.5,meeting the requirements for algorithm deployment in embedded devices with low computational power and memory.(3)In order to realize the automation of garbage sorting,an automatic garbage sorting system under the experimental state is proposed,and the communication mode and software architecture of the system are designed.At the same time,the actual construction of the system is completed,and the automatic garbage sorting is realized,which verifies the feasibility of the algorithm and system proposed in this paper,and provides support for the actual landing of the algorithm.This topic provides intelligent algorithms for garbage sorting automation,and proposes and builds an automatic garbage sorting system,which verifies the feasibility of garbage sorting automation and has important theoretical and practical significance.