Intelligent Automatic Sorting Trash Can Algorithm Design And Embedded Implementation

In recent years,due to the rapid development of society,the amount of domestic waste generated every day is also increasing.At present,incineration is still used for the treatment of domestic garbage,which pollutes the environment and is extremely unfavorable to the recycling of renewable resources.Therefore,starting from July 2019,Shanghai,my country,first carried out waste classification work,and then cities across the country also accelerated the promotion of waste classification systems.However,many troubles have appeared in the process of implementing the waste sorting policy,such as the complexity and variety of waste types,the inadequate monitoring of waste sorting,the lack of knowledge of waste sorting among citizens,and even some citizens who are unwilling to sort waste due to busyness and other reasons.In response to the above problems,this paper combines mechanical design,sensors,Internet of Things,embedded system design,deep learning,machine vision and other technologies to design a smart trash can that can be automatically classified,and proposes a deep residual neural network.The main content of the garbage image classification and recognition method is as follows:First of all,starting from the design of the trash can prototype,the parts and transmissions of the intelligent automatic classification trash can are innovated.In order to achieve the purpose of automatic classification,a camera,stepping motor,steering gear,and infrared are installed on the traditional trash can.And other sensor modules,using a ball screw mechanism to complete the classification and delivery of different types of garbage,and combined with the OneNET Internet of Things platform to enable garbage managers to remotely manage garbage bins.Secondly,based on the TensorFlow framework and Inception V3 deep neural network structure on the GPU,the garbage image data set is migrated and learned,and the obtained PB model is embedded and transplanted to the upper computer.Based on Raspberry Pi using Python language to achieve the systematization and understanding algorithm of garbage images,based on STM32 using the serial port to realize the communication between the upper computer Raspberry Pi and the lower computer STM32.Finally,starting from the theory of algorithm implementation,the original ResNet50 neural network structure is improved,by reducing the size of the convolution kernel,increasing the network width,thereby reducing the model training time and improving the model learning ability.Combining the characteristics of various types and shapes of garbage images,a method of classification and recognition of garbage images based on deep residual neural network is proposed.The test consequence reflect that compared with other neural network structures,the improved ResNet50 structure proposed in this paper can classify and recognize garbage images as high as 95.19%,and it also avoids problems such as the disappearance of network gradients caused by deepening the number of network layers.