热轧钢板表面缺陷模式分类器设计与实现  

Design and Realization of Pattern Classification Sorter for the Surface Defects in Hot-rolling Steel Plate

作　　者：耿瑞芳[1] 曹辉[1] 马永华[1] 

机构地区：[1]北京联合大学生物化学工程学院,北京100023

出　　处：《微计算机信息》2010年第25期188-189,130,131,共4页Control & Automation

摘　　要：热轧钢板是冶金行业的一种重要产品,钢板表面缺陷的计算机自动分类识别系统是当前的研究热点。本文主要针对这一应用背景设计了一种基于BP算法的人工神经网络分类器,用以实现缺陷的分类识别。通过对该神经网络分类器的训练和分类测试,在测试的10个数据中对1、4、5类缺陷(分别对应带状缺陷、凹凸缺陷和裂纹缺陷)识别效果较好;但该网络对缺陷类型2(龟裂)和缺陷类型3(密集的不规则分布缺陷)的识别没有达到预期的效果,还有待于进一步改进。The hot-rolling steel plate is one of important products in metallurgy vocation,the computer automatic classified recognition system on the steel plate surface lacuna is the current research hot spot.The article mainly aimed at this application background to design one kind of sorter based on the BP algorithm artificial neural networks,so that to realized the classified recognition about the lacuna.Through training and the classified test on the neural network sorter,in the 10 testing datas,for 1,4,5 kind of flaws（corresponds belt-shaped flaw,concave-convex flaw and crack flaw separately）,the distinguishing effect to be better,the accuracy rate of recognition is more than 90%;But this network to flaw type 2（chap） and flaw type 3（crowded irregular distribution flaw）,the recognition has not achieved the anticipated effect,and also waits for further improves.

关 键 词：热轧钢板 模式识别 神经网络 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]