Semantic Segmentation of a Printed Circuit Board for Component Recognition Based on Depth Images

• DOI: 10.3390/s20185318
• Publication Date: 2020-09-17T12:29:43Z
• Authors: Dongnian Li; Changming Li; Chengjun Chen; Zhengxu Zhao
• Keywords: PCB; random decision forest; Chemical technology; component recognition; 0202 electrical engineering, electronic engineering, information engineering; depth image; pixel classification; TP1-1185; 02 engineering and technology; semantic segmentation; Article

Locating and identifying the components mounted on a printed circuit board (PCB) based machine vision is an important challenging problem for automated PCB inspection recycling. In this paper, we propose semantic segmentation method depth images that segments recognizes in through pixel classification. The image training set was automatically synthesized with graphic rendering. Based series of concentric circles centered at given pixel, extracted difference features from to train random forest classifier. By using constructed classifier, performed segment recognize Experiments both synthetic real test sets were conducted verify effectiveness proposed method. experimental results demonstrate our can most PCB. Our immune illumination changes be implemented parallel GPU.