Research On Garbage Classification Algorithm Based On Deep Learning

Garbage recycling and reprocessing is always a problem that cannot be ignored in human society.Especially since the beginning of the 21st century,with the rapid increase in human productivity,the amount of waste has increased rapidly.Properly disposing of garbage can make efficient use of resources,reduce pollution,and even create benefits again.In addition,the working environment for garbage disposal is generally harsh.Sometimes the transported garbage is mixed with some glass waste,used batteries,etc.,which will put the garbage sorting workers at a certain risk of injury.Therefore,it is very important to establish a fully automatic intelligent garbage sorting system.This paper designs an image recognit ion and robotic arm intelligent sorting system,and its working process is as follows: The garbage is transported to the conveyor belt.The camera is placed above the conveyor belt.The garbage picture is taken by the camera.After the garbage classificat ion algorithm,the garbage in the picture is identified and located,and the processed information is sent to the back of the robotic arm for grabbing.This can reduce labor costs and improve work efficiency.However,this system puts forward high requirements on the garbage classification algorithm,which requires the algorithm to have a high recognition rate,rapidity and robustness.In recent years,object detection algorithms based on deep learning are in full swing.There are already many excellent target detection networks,but few are used in the field of garbage detection.Therefore,to carry out research on garbage classification algorithms based on deep learning has important theoretical significance and practical value.This article first invest igates the current status of the garbage sorting field,especially the current research status of garbage sorting algorithms and garbage sorting devices at home and abroad,and expounds the design ideas of garbage sorting algorithms;Study the mathematical foundation and basic unit of garbage classification algorithm;The optimization technology of neural network is introduced,and how to prevent over-fitting is explained from the three aspects of data enhancement,dropout technology and normalized data input,which lays the foundation for the design of garbage classification algorithm in the future.Secondly,the design process of the garbage classification network is studied;According to the actual situation,the garbage detection data set is designed;The classic image classification network was studied,and the network with the best classification performance was selected as the baseline;Aiming at the problems of complex background,different shapes,and difficulty in feature extraction of garbage in the sorting site,an attention mechanism is introduced to improve the network’s ability to capture target features;Set up a comparative experiment,studied the influence of the attention module on the garbage classification network,and designed the final HGCNet garbage classification network.Thirdly,the design process of garbage detection and tracking algorithm is studied.Take HGCNet designed in Chapter 3 as the backbone feature extraction network;In view of the difference between the deep feature layer and the shallow feature layer in the generality of information,the feature enhancement and feature pyramid networks are used to fuse the different feature layers,thereby designing a feature extraction network;In view of the large gap between the foreground and background distribution of the network in the training process,the category loss function is optimized and designed;In order to speed up the network convergence,the three factors of the IOU relationship between the detection frame and the real frame,the distance between the center point and the frame aspect ratio are fully considered,and the regression loss function is designed;in accordance with the real-time requirements of the system in actual work,Using the idea of transfer learning,the network is designed to be light-weight;Considering the requirement of continuous tracking of garbage targets in the actual garbage capture process,the multi-sort algorithm and the deep sort algorithm are studied and analyzed;Using the previously designed garbage detection network as a detector,an improved Deep Sort algorithm is designed as the final garbage detection and tracking algorithm.Finally,design and implement the garbage classification system.Investigated the current basic categories of urban garbage,and designed a multi-level garbage classification system according to the characteristics of different garbage;It focuses on the image recognition and robotic arm subsystems,and introduces the hardware selection in detail;The comprehensive monitoring system for waste classification was designed,the design ideas and funct ions of the system were clarified,and each working mode was tested systematically.