摘要
传统带钢生产采用人工目测的方法检测钢板的缺陷类型,检测结果的准确率对检测距离、检测人员的工作经验以及疲劳程度等具有一定的依赖性,这种人工目测方法效率较低。针对传统检测方法中存在的问题,运用支持向量机的优良分类性能,将其应用于带钢表面缺陷的识别研究。实验选用带钢生产现场采集的缺陷图像样本,分别采用支持向量机与决策树算法进行训练与识别,通过实验数据比较发现,支持向量机的sigmoid核与线性核算法要好于决策树算法。在相同的实验环境下,分别采用径向基核,多项式核,sigmoid核以及线性核算法进行训练与识别,通过训练时间的比较可以看出,线性核的训练时间明显较长,径向基核,多项式核与sigmoid核的训练速度相差不大。对比识别率可以发现,径向基核的识别率较好。
In order to solve the problem,this paper took the advantage of support vector machine in classification and applied it to the research on recognition of strip steel surface defect.This paper used support vector machine and decision tree algorithms to train and recognize the images which collected from the work local.The experiment result shows that the sigmoid kernel and linear kernel have better performance than decision tree algorithm.Under the same experimental environment,chose different kernel functions to train and recognize the images.From the comparison of the time and recognition rate,we find that linear kernel needs longer time and the training speed of other three kernels are the same.
出处
《工业控制计算机》
2012年第8期99-101,共3页
Industrial Control Computer
基金
国家自然科学基金(60705012)
湖北省自然科学基金杰出青年人才(2010CDA090)
武汉市科技局晨光计划(201150431095)
关键词
带钢表面缺陷
训练
识别
支持向量机
strip steel surface defect
train
recognition
support vector machine
