基于FasterR-CNN的PCB缺陷检测研究

Research on PCB Defect Detection Based on Faster R-CNN

为了快速准确地进行PCB缺陷检测,文中针对常见的PCB缺陷铜面残渣(简称re)和铜面异物(简称fb),利用了FasterR-CNN进行缺陷目标检测。从测试结果可以看出,FasterR-CNN检测模型在PCB缺陷检测中具有良好的检测效果。当阈值为0.5时,有缺陷图片(简称NG图片)的漏检率只有1.73%,无缺陷图片(简称OK图片)的误检率只有5.20%,基本满足了本文的PCB缺陷检测要求。

In order to accurately and quickly detect the PCB defects, the Faster R-CNN is used to detect the common PCB defects, such as the copper surface residues(referred to as re) and copper surface foreign objects(referred to as fb).It can be seen from the test results that the Faster R-CNN detection model has a good detection effect in PCB defect detection.When the threshold is 0.5, the missed detection rate of defective images(referred to as NG images) is only 1.73%, and the false detection rate of non-defect images(referred to as OK images) is only 5.20%.The result basically meets the PCB defect detection requirements.