Intelligent Garbage Fine Sorting System Of Small And Array Assembly Line

In recent years,the problem of "garbage siege" in China has become more and more serious.Some large and medium-sized cities have begun to implement the regulations on classification of domestic garbage compulsorily.However,the national awareness of garbage classification is generally weak and their classification knowledge is insufficient,so the implementation of garbage classification has made slow progress and the recovery rate of waste resources is low.Therefore,it is very necessary to design an intelligent waste fine sorting system to improve the garbage recycling,reduction and harmless ratio.This thesis plans to develop an assembly line intelligent sorting system applied to small waste collection stations,focusing on solving the problems such as the effective utilization of a large number of unlabeled garbage image data,high-precision multi-object detection on waste assembly line and the design of efficient and economical waste intelligent fine sorting system.The main work of this thesis is as follows:In order to effectively utilize of a large number of unlabeled garbage images,a high-quality labeled dataset is collected and constructed,in which one image contains one garbage object,called Garbage Net.This dataset contains more than 50000 garbage images.Besides,more than 170000 low-quality unlabeled single-object garbage images are collected from internet.Finally,a semi-supervised single-object garbage image dataset,S-Garbage Net,is established.Then,an improved Fix Match single-object garbage image classification algorithm is proposed by comprehensively application of positive and negative pseudo label,the consistency regularization of strong and weak data augmentation and the update of model weights by exponential moving average.The results on S-Garbage Net dataset show that the semi-supervised method proposed in this thesis is significantly better than the supervised Image Net transfer learning method in lack of labeled image data.In order to achieve high-precision multi-object detection on garbage assembly line,the garbage object is segmented by the salient object detection algorithm in the singleobject garbage dataset Garbage Net.Then a realistic object detection dataset MGarbage Net,containing 30000 multi-object garbage images,is constructed by randomly combines the garbage object.Next,aiming at adapting the special scene of garbage multi-object detection better,an improved Faster RCNN garbage multi-object detection algorithm,which baseline is Faster RCNN,is designed by comprehensively using deformable convolution,PAFPN,mixup,cutout,color jitter,multi-scale training,S-Garbage Net transfer learning.Finally,this algorithm is tested on the M-Garbage Net dataset and its MAP@0.5 reaches 88.0%,which is 8.2% higher than the classical Faster RCNN algorithm.An efficient and economical small array assembly line intelligent fine sorting system is designed.The main mechanical structure of the sorting system is composed of a garbage belt delivery platform,an array sliding table and a garbage grabbing unit.The control system is mainly composed of PLC control system and visual detection system.Finally,the intelligent waste fine sorting task is completed through visual multiobject detection,the optimal allocation of grabbing task and combined with the coordination of various executive parts.