Public Garbage Classification System Based On MobileNetV3

In the field of environmental protection,the effective classification of domestic waste has gradually become the focus of social development,and the classification process of domestic waste is the key to the treatment of domestic waste.Under the current situation of our country,the waste classification and treatment in the waste treatment plant mostly adopts the way of manual assembly line sorting,which has the disadvantages of bad environment,high labor intensity,low sorting efficiency and weak automation,far from meeting the needs of the development and social progress of environmental protection resource recycling and utilization in our country.With the continuous improvement of China’s industrial automation level,the automatic production mode has been well applied in many industries.However,in the environmental protection industry,it is restricted by factors such as late start,less investment,and difficult sorting.The automatic sorting equipment suitable for China’s national conditions is scarce.Therefore,in view of the waste sorting demand of the environmental protection industry,the scheme of replacing manual sorting with automatic industrial equipment is proposed It is imperative to develop an intelligent classification system for domestic waste.Garbage classification is the first step of garbage separation,recycling or reuse.In this paper,we use the deep learning classification model based on MobileNetv3,which classifies the common garbage according to the following categories: recyclable,kitchen waste,hazardous waste and other garbage.15835 garbage image data sets in.JPG format are used for training.This model uses the model trained on the large-scale visual recognition challenge dataset of Imagenet for migration learning.The baseline mobilenet V3 model is optimized,and the final test accuracy is 78%.The model is more suitable for mobile device deployment.This paper also uses docker integrated deployment model training environment,and optimizes the algorithm to build an efficient neural network.Finally,the trained model is deployed in the front-end equipment to make full use of the edge computing resources,solve the problem of information delay in the engineering scene,and achieve the engineering purpose of real-time data processing.Complete a detailed engineering process and algorithm verification.