Deep learning-enhanced defects detection for printed circuit boards

Printed circuit boards (PCBs) are an important component of electronic devices. Therefore, ensuring the quality of such PCBs in the manufacturing process is crucial. Especially, cracks or scratches appearing on the PCB surface pose a significant hurdle, due to their minuscule size, making them the most challenging to address. In this work, we present a real-time automated algorithm for defects inspection of printed circuit boards (PCBs) in different lighting conditions. First, the Oriented FAST and Rotated BRIEF (ORB) algorithm extracts features from the input images, then the Brute-force matching method matches these features with the ORB features template. Next, the input images are calibrated to match the size and orientation of the template data by the RANSAC(Random Sample Consensus) algorithm. Finally, the defective areas on the PCB surface are segmented by using the U-NET (i.e., a type of convolutional neural network (CNN)) model. The proposed algorithm is tested in three different lighting conditions: low light, normal light, and high light conditions. Experimental studies are conducted on a representative PCB to evaluate the defect detection capacity of the proposed algorithm and the experimental results show that the proposed system works well in the three different lighting conditions with an accuracy of up to 97%, the detection speed is 12 frames per second (FPS).