Automatic Optical Inspection for Defective PCB Detection Using Transfer Learning

Automated Optical Inspection (AOI), based on machine vision, has been widely used in the industry to perform quality control in various manufacturing segments. In addition to electrical testing, optical analysis is an important step for quality control in Printed Circuit Board (PCB) manufacturing. Two approaches are reported in the literature for the inspection of printed circuit boards: the referential and the non-referential approach. Recently the DeepPCB database, composed of PCB images for identification and classification of defects, was published. This work assess the application of transfer learning strategies to detect defective PCBs in a non-referential method. Since the DeepPCB dataset is too small to train a deep model from scratch, we evaluated some strategies of transfer learning from pre-trained models VGG16 and ResNet50. Given that the PCB images from DeepPCB does not present the same complexity of the ImageNet dataset, feature extraction from a less deep model as VGG16 presented better results. The best-evaluated model obtained an accuracy of 89%.