Defect Detection in Printed Circuit Board Manufacturing Using Part Image Classification Based Whole Image Understanding with Direct Defect Detection

Nowadays, Printed Circuit Board (PCB) is the most important part in the electronics product and any defect occurs in the PCB makes the final product useless. So, PCB inspection in current manufacturing process becomes a crucial part in electronic products. Many PCB industries has developed some automated detection methods in manufacturing process and also traditional PCB inspection was generally supervised by the manual inspection. Because of manual inspection, several issues occurred such as error-prone, time-consuming as well as causing work fatigue. To overcome these challenges, a Part Image Classification based Whole Image Understanding with Direct Defect Detection (PIC-WIU-DDD) is proposed using PCB Boards Defect Detection (PCB-BDD) dataset. For image preprocessing, z-score normalization and Gaussian filter are used. Next, Squeeze and Excitation (SE) module and Convolutional Block Attention Module (CBAM) are used for feature extraction process which increases accuracy also reduces false positives. Further, for classification purpose PIC-WIU-DDD techniques are used to reduce computational complexity, improve robustness and reduce false positives. The results prove that the proposed PIC-WIU-DDD show better performance in terms of accuracy, precision and F1 score when compared to previous existing model like Convolutional Neural Network (CNN).