Research On Fine-grained Garbage Classification Based On Deep Learning

Nowadays,garbage classification has become a hot topic in society.In fact,on April 26,2019,the Ministry of Housing and Urban-Rural Development of my country and other departments issued the "Notice on Comprehensively Carrying out Domestic Waste Classification in Cities at the Prefectural Level and Above in the Country",and decided to start at the prefectural level and above in 2019.The city has started the classification of domestic waste in an all-round way.By the end of 2020,46 key cities have basically completed domestic waste classification and treatment systems.Manual garbage sorting is the first link in garbage disposal,but the link that can handle massive amounts of garbage is the garbage treatment plant.However,the current domestic garbage treatment plants basically use manual assembly line sorting for garbage sorting,which has disadvantages such as harsh working environment,high labor intensity,and low sorting efficiency.In the face of massive garbage,manual sorting can only sort out a very limited part of recyclable garbage and hazardous garbage,and most garbage can only be landfilled,which brings great waste of resources and danger of environmental pollution.With the application and development of deep learning technology in the field of vision,we have seen the possibility of using AI to automatically classify garbage,taking garbage pictures through sensors,and using deep neural classification networks to detect the types of garbage in the pictures,so that the machine can automatically Carrying out fine-grained sorting of garbage,greatly improving the efficiency of garbage sorting.Fine-grained garbage classification is to determine whether there is garbage in a given image.Fine-grained garbage is mainly identified by the detailed classification of garbage,as an alternative to the current high-cost manual classification,or used for smart trash cans.The main research contents of this paper include:(1)Perform feature selection based on existing data sets.By observing the characteristics of various types of garbage and analyzing it,a suitable preprocessing method for the input image is researched.(2)Research a variety of data enhancement methods,perform data enhancement,cut and flip the existing image data,and enhance the generalization ability of the model.(3)Research a variety of deep neural classification networks,and improve the neural network structure.Through the research and analysis of the Res Net network structure,the Res Ne Xt network structure and the Efficient Net network structure,comparative experiments are carried out,and the Efficient Net-B5 basic network model suitable for fine-grained garbage classification is selected,and then the network structure of Efficient Net-B5 is adjusted and optimized,Add attention mechanism,optimization of loss function and optimizer,use label smoothing for sample imbalance problem,carry out migration learning and other methods to meet the performance requirements of fine-grained garbage classification.(4)Completed the research of fine-grained garbage classification based on deep learning,realized high-accuracy fine-grained garbage classification,and improved the accuracy and speed of classification.