基于机器视觉的智能垃圾分类实验平台设计  

Design of Intelligent Garbage Classification Experimental Platform Based on Machine Vision

作　　者：李祺 曾明[1] 卢向哲 聂为之 LI Qi;ZENG Ming;LU Xiangzhe;NIE Weizhi(School of Electrical Engineering and Automation,Tianjin University,Tianjin 300072,China;National Virtual Simulation Experimental Teaching Center of Electrical Engineering and Automation,Tianjin University,Tianjin 300072,China)

机构地区：[1]天津大学电气自动化与信息工程学院,天津300072 [2]天津大学电气工程与自动化国家级虚拟仿真实验教学中心,天津300072

出　　处：《实验室研究与探索》2022年第4期68-73,共6页Research and Exploration In Laboratory

基　　金：国家自然科学基金项目(61872267);天津大学实验室建设与管理改革项目(LAB2021-09);天津大学研究生创新人才培养项目(YCX202037)。

摘　　要：针对前端垃圾分类采集的需求,研发了基于机器视觉的智能垃圾分类实验平台。该平台使用“视觉检测与智能桶分离”及“桶盖与桶身分离”的机构设计,采用树莓派CM4、STM32F103控制板及云平台组建云边协同的控制系统。通过构建大规模单物品垃圾图像数据集GarbageNet,设计基于深度跨连接网络的垃圾分类算法,实验平台对10大类垃圾识别率达97.2%。同时,面向实验教学需求,开发了通用图像分类模型实验架构,并设计了基于迁移学习的实验内容,在实验教学过程中取得良好成效。In view of the requirement for garbage classification,an intelligent garbage classification experimental platform was developed.It adopts visual inspection module and intelligent garbage which can separate the lid and can.The experimental platform has a cloud-edge collaboration control system,which was comprised by Raspberry PiCM4,STM32 F103 controller and cloud platform.By using the large-scale image data set GarbageNet and the garbage classification algorithm based on deep skip connection network,10 categories garbage recognition rate of the experimental platform achieves 97.2%.Depending on the experimental platform,experimental architecture of universal image classification model and experiment content based on transfer learning are developed for practical teaching.The platform has been applied to experimental teaching,which enriched teaching methods and achieved good results.

关 键 词：垃圾分类 迁移学习 图像数据集 智能垃圾桶 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]