基于改进Inception ResNet V2网络的可回收垃圾分拣系统  

A RECYCLABLE WASTE SORTING SYSTEM BASED ON AN IMPROVED INCEPTION RESNET V2 NETWORK

作　　者：徐丽 周腊吾[1] 李高嘉 XU Li;ZHOU Lawu;LI Gaojia(School of Electrical and Information Engineering,Changsha University of Technology,Changsha 410114,China)

机构地区：[1]长沙理工大学电气与信息工程学院,长沙410114

出　　处：《环境工程》2024年第4期233-241,共9页Environmental Engineering

基　　金：湖南省自然科学基金项目(2018JJ2434);长沙理工大学研究生科研创新项目(CSLGCX23059)。

摘　　要：垃圾围城一直是困扰我国城市管理的一大难题。针对可回收垃圾处理过程难度大的问题,提出了1套基于改进的Inception ResNet V2网络结合ROBOT MG400机械臂进行可回收垃圾自动化分拣的系统。首先,对MG400机械臂上的夹具进行改进,使之更适用于垃圾抓取;然后,自主创建了50850张数据集,在此基础上对垃圾图像经过背景降噪、图像分类以及投票算法的处理,并在Inception ResNet V2网络的输出层加入CBAM注意力机制,提高模型识别的准确性;最后对整个系统进行了实验验证。结果表明:该系统可较为准确地分类垃圾并收集至对应垃圾收集容器中,训练时模型的准确率为99.35%,在系统中传送带运行时识别准确率为95.39%,改进的网络在实际应用中的mAP值比原模型高2.56%,并且系统的分拣效率可达到60件/min。该系统可高效率、高准确率、高精度独立地完成可回收垃圾的分拣工作。Garbage siege has always been a big problem in China’s urban management.Aiming at the difficulties of the recyclable waste disposal process,this paper proposed a system based on the improved Inception ResNet V2 network combined with the ROBOT MG400 robotic arm for automatic sorting of recyclable waste.Firstly,we improved the fixture on the MG400 robotic arm to make it more suitable for garbage grabbing.Then,we independently created a dataset of 50850 sheets,based on which the garbage images were processed by background noise reduction,image classification,and voting algorithms,and the CBAM attention mechanism was added to the output layer of the Inception ResNet V2 network to improve the accuracy of model recognition.Finally,the whole system was experimentally verified.The verification results showed that the system could classify garbage more accurately and collect it into the corresponding garbage collection container,and the recognition accuracy was 99.35%and 95.39%when the conveyor belt was stationary and running,the mAP value of the improved network was 2.56%higher than that of the original model,and the sorting efficiency of the system reached 60 pieces per minute.Therefore,this system can independently complete the sorting of recyclable waste with high efficiency,high accuracy,and high precision.

关 键 词：Inception ResNet V2 图像分类 可回收垃圾 深度学习 注意力机制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术] X799.3[环境科学与工程—环境工程]