摘要
为了有效提高在线带钢表面缺陷检测的识别率和实时性,提出了一种优化的量子粒子群-径向基函数(QPSO-RBF)网络的带钢缺陷分类识别算法。首先采用加权模糊C-均值聚类(WFCM)算法确定RBF网络隐含层参数,算法对带钢缺陷特征数据出现的团状分布与疏散分布问题能够达到很好的聚类划分,避免对特征数据集等划分的趋势;然后采用QPSO算法对RBF网络的参数编码成粒子个体,在全局空间中动态地搜索最优适应值的RBF网络参数,提高了网络的学习性能,并建立了带钢缺陷分类识别的专家知识库。实验结果表明:本文算法可以自动获得较优的网络结构,收敛速度快,对带钢缺陷的平均识别率为94.63%,平均误识率为3.0%,对测试样本的识别时间为4ms,小于生产线上每张图片的采集周期10ms,因此,可以为高速生产线上的带钢表面缺陷在线实时检测提供了有利条件。
In order to improve the detection of surface defects of the striprs precision and real-time per- formance,a new optimized QPSO_RBF network is used in strip defect classification and recognition. Firstly, the parameters of RBF hidden layer are determined by using weighted fuzzy C-means （WFCM） algorithm, cluster distribution and evacuation distribution of the strip feature data can be well handled hy WFCM algorithm. The algorithm can avoid feature data set equal partition trend, Then,all network pa- rameters are coded to individual particles in this algorithm, the parameters can dynamicaly search opti- mal-adaptive values in global space by quantum particle swarms optimization （QPSO）, the performance of network learning is improved,and the strip defect classification and recognition expert knowledge base is established. The experimental results show that the algorithm can obtain more excellent network structure,efficient convergence, the average recognition rate for strip defects is 94. 63%, the average course rate is 3.0 %, the recognition of test sample time is 4 ms, and the recognition of test sample time is 10 rns less than the each image collection cycle on production. So the algorithm of this paper can pro- vide favorable conditions for the scene of the high speed production line of steel strip surface defect real- time detection.
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2015年第2期320-327,共8页
Journal of Optoelectronics·Laser
关键词
带钢表面缺陷
实时检测
特征提取
分类识别
surface defect of strip
real-time detection
feature extraction
classification and recognition
