基于卷积神经网络的零件识别与分拣系统  

Parts Identification and Sorting System Based on Convolution Neural Network

作　　者：周玉强 陈国栋 盛小明[1] Zhou Yuqiang;Chen Guodong;Sheng Xiaoming(College of Mechanical and Electrical Engineering, Soochow University, Suzhou 215006, China)

机构地区：[1]苏州大学机电工程学院,江苏苏州215006

出　　处：《煤矿机械》2018年第6期151-153,共3页Coal Mine Machinery

基　　金：国家自然科学基金项目(U1509202)

摘　　要：针对工业领域的生产线零件分拣系统存在识别零件准确率较低的问题,在对原始图片进行背景减除、滤波、裁剪零件区域和尺寸调整的基础上,提出利用卷积神经网络,进行零件的识别分类。研究了卷积神经网络对各种姿态和不同光照条件下的零件识别准确率,同时搭建了一套机器人系统实现零件分拣任务。实验表明,卷积神经网络可以准确地识别各种零件,且对光照和姿态有较强的鲁棒性,对生产线零件分拣系统有一定的应用价值。To solve the problem of low-accuracy in parts identification existing in the production line parts sorting system in the industrial field,putting forward the identification and classification of parts by using Convolution Neural Network（CNN）,based on the background subtracting,filtering, clipping parts zone and size adjustment of the original image. Studying the parts recognition accuracy of CNN for various positions and under different illumination conditions,and a robot system is constructed to complete the parts sorting task. The experiment shows that CNN can accurately identify various parts,and has strong robustness for illumination and positions. The results have a certain application value to the production line parts sorting system.

关 键 词：卷积神经网络 分拣系统 图像处理 图像识别 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]