基于图像处理和支持向量机的微型齿轮缺陷检测  

Defect detection for microgear based on image processing and support vector machine

作　　者：贺秋伟[1] 王龙山[1] 于忠党[1] 李国发[1] 高立国[1] 

机构地区：[1]吉林大学机械科学与工程学院,长春130022

出　　处：《吉林大学学报（工学版）》2008年第3期565-569,共5页Journal of Jilin University:Engineering and Technology Edition

基　　金：吉林省科技发展计划项目(20040534)

摘　　要：针对微型齿轮缺陷传统检测手段落后、准确率低、不易在线实施、受人为因素影响等问题,提出了以电荷耦合器件为图像传感器,采用图像处理技术和支持向量机对齿轮缺陷进行检测的方法。首先,系统采用发光二极管照明光源提供高强度背光照明,使用A102FCCD数字摄像头采集齿轮的图像,经过图像采集卡传输到计算机。其次,采用边缘保持滤波器对含有噪声的原始数字图像进行降噪处理,采用迭代阈值法和Otsu双阈值法对齿轮图像进行分割,形成二值化图像。然后获取齿轮样本,提取样本特征。最后用支持向量机来构造齿轮缺陷识别模型。该方法识别正确率达97.8%。理论分析及实验结果表明,该方法检测成本低廉、可靠性高、推广性强、容易在线实施。To get rid of the weaknesses of the traditional defect detection of the microgear, such as low accuracy, inconvenient on-line implementation, and liable to man-made interference, a new detection technique for microgear defect based on the image processing technology using the charge coupled device（CCD） as the image sensor and the support vector machine（SVM） was proposed. The brilliant back illumination was provided by a light emitling diode（LED） illuminating source, the digital image of the tested microgear was collected by an A102FCCD digital camera and input into a computer by an image acquisition card. The original gray level digital image with noise was processed by the edge retention filter to reduce its noise, and segmented by the iteration threshold method and the Otsu dual threshold method, and the binary image was formed. The samples of the gear were collected and the sample features were extracted. A gear defect recognition model was built based on the SVM. The recognition accuracy is as high as 97.8%. The proposed technique is characterized by low cost, high relialility, good popularization, and easy on-line implementation.

关 键 词：机械制造自动化 图像处理 缺陷检测 电荷耦合器件 支持向量机 微型齿轮 

分 类 号：TH132[机械工程—机械制造及自动化] TP274.5[自动化与计算机技术—检测技术与自动化装置]