Automatic visual inspection for printed circuit board via novel Mask R-CNN in smart city applications. (April 2021)

• Record Type: Journal Article
• Authors: Lian, Jian; Wang, Letian; Liu, Tianyu; Ding, Xia; Yu, Zhiguo
• Is Part Of: Sustainable energy technologies and assessments. Volume 44(2021)
• Publication Date: 2021-04
• Subjects: Machine learning -- Machine vision -- Automated visual inspection -- Printed circuit board; 00-01 -- 99-00; Renewable energy sources -- Periodicals; Energy development -- Technological innovations -- Periodicals; Electric power production -- Periodicals; Energy storage -- Periodicals; 333.79
• DOI: 10.1016/j.seta.2021.101032
• Languages: English
• ISSNs: 2213-1388

Abstract

Abstract: The increasing population in the whole world demands adequate infrastructure to satisfy varied requirements. To fulfill this requirement, the introduction of information techniques presents an opportunity for the development of smart cities. For instance, an automatic visual inspection can be employed to replace the role of workers in quality management and streamlines automation. Previously, a large amount of machine vision-based algorithms has been presented to address this problem. However, accurate detection of various tiny integrated circuits remains an unresolved issue. To bridged this gap, a novel deep learning-based approach was proposed for instance segmentation in printed circuit board images. By adding the geometric attention-guided mask branch into the fully convolutional one-stage object detector under the framework of Mask R-CNN, it can produce a segmentation mask for each bounding box to enhance the identification accuracy. To evaluate the ability of the proposed approach, the comparison experiments were conducted between state-of-the-art techniques and ours. Experimental results demonstrate that the presented algorithm outperformed the state-of-the-art both in precision, sensitivity, and accuracy for both small devices like resistors and capacitors as well as integrated circuits.