Hybrid Convolutional Neural Network with Transfer Learning and MobileNetV2 for Advanced Multi-Class PCB Defect Detection

Printed Circuit Boards (PCBs) are integral to modern electronic devices, but their manufacturing processes are prone to defects that can compromise quality and reliability. This study introduces an enhanced hybrid convolutional neural network (CNN) model leveraging transfer learning and MobileNetV2 to achieve efficient and accurate detection of six common PCB defects: Missing_hole, Mouse_bite, Open_circuit, Short, Spur, and Spurious_copper. By incorporating advanced image preprocessing, data augmentation techniques, and a customized fully connected layer, the proposed model effectively addresses challenges in defect detection. Experimental evaluations demonstrate a classification accuracy of 96% on the test set, with ROC curves achieving AUC values exceeding 0.99 for all defect categories. Comparative analyses reveal significant performance improvements over traditional methods, including a backpropagation neural network (BP NN), underscoring the model's ability to generalize across diverse and complex defect patterns.