A VGG-16 based Faster RCNN Model for PCB Error Inspection in Industrial AOI Applications

Abstract

To detect product error and modify the product error, most industry are using human eyes. However, it is not only costs time but also costs money. Our purpose is to develop a model to detect the PCB board errors and draw the bounding boxes. The model is going to be developed with a pre-trained model VGG16 and data collected from Adventech corp. The error types of training data have been speared into five error types (Bridge, Appearance, Empty, Solder_ball, Solder_balls), where the highest AP result of these classes is over 90%.