Convolutional Neural Network-based Research And Application For Garbage Detection

Since the General Office of the People’s Republic of China promulgated the notice of domestic waste classification,the topic of waste classification has once been hotly discussed on the Internet.Garbage classification is still a relatively trendy lifestyle for Chinese residents.However,my country’s population base is large and widely distributed,and the increasing ageing of the population has blocked the development of garbage classification.The garbage classification text query system promoted can solve most residents’ doubts about garbage classification.However,there are still a large number of elderly and other groups without independent behavior ability who lack the ability of handwriting,pinyin or voice input in my country.Therefore,the use of computer vision,which is intuitive,simple,and low in learning threshold,to achieve the auxiliary role of the garbage classification system has more universal applicability and application value.This paper combines the problem of garbage classification with deep learning technology,uses convolutional neural networks to locate and classify garbage in the image,and implants the system into mobile terminal equipment to assist Chinese residents in the garbage classification work.The main research contents of this paper are as follows:(1)In response to the current scarcity of data sets in the field of garbage detection,this paper organizes,selects and expands garbage images on the network,provides target positioning and classification and labeling,and also performs data enhancement processing to expand training samples,and establishes an application that can be applied The garbage image data set for the object detection task.(2)Proposed a lightweight garbag detection network based on Mobile Net-SSD,Mobile Net V3 replaces the backbone network VGG16 in SSD,and the network structure is fine-tuned to adapt to garbage detection tasks.The network is a multi-task framework,which completes the two tasks of confidence box regression and target classification.(3)Research and analysis of mainstream object detection models such as Faster R-CNN,YOLOv3,SSD,and compare them with Mobile Net-SSD from multiple perspectives such as m AP,FPS,model size,and parameter amount.In the application scenario of garbage detection,the lightweight convolutional neural network Mobile Net-SSD achieves a 97.9% increase in detection speed and a 62.7% reduction in model size under the premise that m AP is only 0.35% lower than the original SSD model.(4)Developed a domestic garbage detection system based on Mobile Net-SSD,including registration,login,password modification,garbage image detection,and query detection history functions to complete the task of using mobile devices to detect garbage in garbage images.