摘要
为了实现在工业环境下的织物瑕疵在线检测,提出了一种基于单类支持向量机(OCSVM)的织物异常纹理检测方法。通过利用CCD采集织物图像,滤除图像噪声后提取了图像小区域窗口子图像特征;通过实验寻找了两组有效的特征向量,对特征值进行了归一化和主成份分析降维后导入支持向量机分类器中进行了训练,利用单类SVM对异常区域进行了定位和标记。通过对分别利用两组特征向量识别出的图像结果进行组合得到了最后的瑕疵区域。实验结果表明,该算法能够正确地对多种瑕疵进行识别,并能较大程度降低误检率和漏检率;同时,能够有效解决生产实际中瑕疵训练样本难以获取的问题,对未知的待测样本有较好的推广性,可以适应工业检测的要求。
In order to realize on-line defect detection of fabric in real industry,an abnormal fabric detection method based on One-Class Support Vector Machine( OCSVM) was proposed. The fabric images were collected by a CCD camera and were filtered by median filter before the features of sub-images were extracted from the divided rectangular areas. Two groups of effective feature vectors were decided by experiments. After the normalization and dimensionality reduction by using principal component analysis,the features were employed in the training of OCSVM,which subsequently could be used to locate and label the abnormal regions. The defective regions could be obtained through the combination of detection results obtained from the two different groups of feature factors. The experimental results indicate that the algorithm could correctly identify different defects and could effectively reduce the false alarm rate and missed detection rate. It provided an available solution to solve the problem of the difficulty in acquiring enough defective samples in the practical production and has a good generalization performance to the unknown test samples. The algorithm can meet the demand of industrial application.
出处
《机电工程》
CAS
2016年第2期237-241,共5页
Journal of Mechanical & Electrical Engineering
基金
国家自然科学基金资助项目(51005077)
教育部高学校博士学科点科研基金(博导类
20133514110008)
国家卫生和计划生育委员会科研基金(WKJ-FJ-27)
国家质检总局科技计划项目(2011QK216)
福建省杰出青年基金滚动项目(2014J07007)
福建省质量技术监督局科技计划项目(FJQI2014008
FJQI2013024)
福建省高等学校学科带头人培养计划(闽教人〔2013〕71号)
福建省自然科学基金项目(2015J01234)
关键词
织物
瑕疵检测
机器学习
支持向量机
fabric
defect detection
machine learning
support vector machine(SVM)