论文部分内容阅读
针对具有小样本、非线性、高维模式识别特点的冷轧带钢表面缺陷,且部分缺陷分布零散、不相连而导致后期识别数量增加、识别率低的情况,提出聚类与优化支持向量机相结合的改进分类算法。利用矩形框将缺陷进行标记,实现缺陷聚类合并,减少后期缺陷识别分类个数,便于后期正确识别判断;利用粒子群优化算法结合交叉验证自动选取最优参数,确定支持向量机结构,并结合实际生产线上出现频率较高的5类带钢缺陷进行分类研究。实验结果表明,相较于改进BP神经网络和网格优化的支持向量机,聚类与优化支持向量机相结合的改进分类算法不仅解决了位置接近的同种缺陷重复分类的问题,而且耗时短、缺陷正确识别率可达98%,符合实际生产线需求。
Aiming at the surface defects of cold-rolled strip with small sample, non-linear and high-dimensional pattern recognition, and the distribution of some defects is fragmented and disjoint, the number of post-recognition increases and the recognition rate is low, the clustering and optimization support vector machine Combination of improved classification algorithm. Rectangular frames are used to mark the defects to realize the defect clustering and merging, so as to reduce the number of late-stage defect recognition and classification, which is convenient for later identification. By using particle swarm optimization algorithm and cross-validation, the optimal parameters are automatically selected and the support vector machine In the actual production line appear higher frequency of 5 types of strip defects classification research. Experimental results show that compared with improved support vector machines based on BP neural network and grid optimization, the improved classification algorithm based on clustering and optimized support vector machines not only solve the problem of repeated classification of the same kind of defects in close proximity, but also time-consuming Short, defective recognition rate up to 98%, in line with the actual production line needs.