Fast SMDs Segmentation on PCB using CNN

Abstract

We propose a deep neural network for PCB inspection. Our network segments 36 types of SMDs from PCB images. We designed the network inspired by the Deeplab-largeFOV. We replace a part of Deeplab-largeFOV with a part of Alexnet to balance the accuracy and speed. Our network achieves 81.9%mIOU accuracy and 10.5FPS speed with a 1024×1024 size image. This is two times faster than the previous method.