摘要
研究提出了新的玻璃制品智能检测系统和算法;根据玻璃制品检测的需要,设计了一个机器视觉检测系统,并开发了实验样机;在获取玻璃制品图像后,根据缺陷的特点来分割出可能缺陷区域,然后在可能缺陷区域内提取缺陷特征;提出采用一种多核函数支持向量机集成方法来对特征进行分类;此多核函数支持向量机集成采用遗传算法来协同优化集成中支持向量机的各项参数,使得各支持向量机在拥有较高分类性能的同时保持差异性;而在最后集成各支持向量机时采用了遗传选择集成方法;实验表明采用文中提出的检测算法在实验样机上检测玻璃制品质量,准确率可达97%以上。
This paper proposes a novel intelligent system and method to inspect glass products. According to requirements of glass products inspection, the inspection system based on machine vision is designed, and the prototype is developed. After capturing the images of glass product, the possible defect regions are segmented by the characteristics of defects, and then the features are extracted in these possible defect regions. The support vector machines ensemble using multi kernel function is put forward as the classifier for classifying the features. The genetic algorithm is used to optimize cooperatively the parameters of the support vector machine in ensemble. This method can make the classifying performance of the SVMs in ensemble is good and the variance of the SVMs in ensemble is big. The selective ensemble method based GA is adopted to ensemble the SVMs. The experiments show that using this method to inspect glass products on the prototype, the accuracy rates reach above 97%.
出处
《计算机测量与控制》
CSCD
北大核心
2011年第2期359-362,共4页
Computer Measurement &Control
基金
国家自然科学基金(U0735003
60604006)
教育部博士点基金(20070562005)
广东省自然科学基金(07117423)
关键词
智能检测
机器视觉
支持向量机集成
遗传算法
intelligent inspection
machine vision
support vector machine ensemble
genetic algorithm.
