多尺度特征PCB裸板缺陷识别  

PCB bare board defect identification based on multi-scale features

作　　者：李任鹏 李云峰[1] LI Renpeng;LI Yunfeng(School of Mechatronics Engineering,Henan University of Science and Technology,Luoyang 471003,China)

机构地区：[1]河南科技大学机电工程学院,河南洛阳471003

出　　处：《中国测试》2024年第9期181-190,200,共11页China Measurement & Test

摘　　要：针对印刷电路板裸板缺陷视觉检测,提出一种基于多尺度特征的缺陷识别算法。首先,为减弱噪声干扰同时保留足够的边缘细节信息,对图像进行高斯滤波和对比度增强,使用多层感知机网络分割出感兴趣区域,然后使用Harris角点检测算法完成图像配准,提取缺陷区域,最后通过分析缺陷区域的多尺度信息,构建高斯金字塔图像,提取多尺度方向投影特征,多尺度共生矩阵特征和多尺度边缘特征,获得缺陷特征描述子,根据特征描述子构建特征向量,利用支持向量机分类器完成对缺陷的分类识别。实验结果表明:所提算法能够对印刷电路板裸板存在的漏孔、鼠咬、断路、短路、毛刺缺陷位置进行定位并精确识别缺陷类型,在小样本场景下识别准确率接近99%。A multi-scale feature-based defect recognition algorithm is proposed for visual inspection of bare PCB defects.First,in order to reduce noise interference while retaining sufficient edge detail information,Gaussian filtering and contrast enhancement are applied to the image,and the region of interest is segmented using a multilayer perceptron network.Finally,by analyzing the multi-scale information of the defect area,constructing a Gaussian pyramid image,extracting multi-scale directional projection features,multi-scale symbiotic matrix features and multi-scale edge features,obtaining defect feature descriptors,constructing feature vectors according to the feature descriptors,and using support vector machine classifier to complete the classification and recognition of defects.The experimental results show that the proposed algorithm can locate and accurately identify the defect types of leakage holes,mouse bites,broken circuits,short circuits,and burr defects existing in bare PCBs with an accuracy of nearly 99%,in small sample scenarios.

关 键 词：图像处理 缺陷检测 多尺度特征 支持向量机 多层感知机 

分 类 号：TB9[一般工业技术—计量学] TN41[机械工程—测试计量技术及仪器]