基于多尺度特征融合的PCB缺陷检测

PCB Defect Detection Based on Multi-scale Feature Fusion

高精度的缺陷检测对印刷电路板制造具有重要意义。基于机器视觉的缺陷检测技术被大量应用于工业生产领域。在生产过程中,对产品进行高度精确、非接触式缺陷检测。论文提出了一种基于多尺度特征融合的印刷电路板缺陷检测神经网络模型。首先,通过传统的图像处理技术,对PCB板缺陷进行预提取。其次,将缺陷预提取图像和PCB模板图像分别进行特征提取。然后将空间金字塔池化网络嵌入到各卷积模块的特征图中,融合多尺度特征向量。最后,利用训练过程中的对比损失,得到模板和待测PCB板相似度度量,从而对缺陷进行精确检测和定位。实验结果表明,该模型与传统的缺陷检测方法相比在检测和定位缺陷,都具有更好的性能表现。

High precision defect detection is very important for PCB manufacturing. The defect detection technology based on machine vision has been widely used in industrial production. During production,highly accurate,non-contact defect detection is performed on the product. A neural network model for PCB defect detection based on multi-scale feature fusion is presented in this paper. Firstly,the defects of PCB board are pre-extracted by traditional image processing technology. Secondly,the defect pre-extraction image and PCB template image are extracted respectively. Then the spatial pyramidal pooling network is embedded into the feature map of each convolution module to fuse multi-scale feature vectors. Finally,the similarity measurement between the template and the PCB to be tested is obtained by using the comparison loss in the training process,so as to accurately detect and locate the defects. The experimental results show that this model has better performance in detecting and locating defects than traditional defect detection methods.