PCB Assembly Component Recognition based on Semantic Segmentation and Attention Mechanism

Xingyao Huang ； Taifu Li ； Wenbin Chen

《International Core Journal of Engineering》 8卷4期 (2022/04) Pp. 299-310

https://doi.org/10.6919/ICJE.202204_8(4).0039

摘要

Focusing on the current online inspection of PCB assembly process, traditional image inspection methods cannot extract high-level image features and high noise sensitivity. Our purpose is to use the semantic segmentation method to extract the high-level semantic features of PCB assembly images, realize the outline segmentation and identification of components, and explore an innovative PCB component detection method. The implementation method is to improve Deeplab v3+ by adding attention mechanism and multi-scale input channels, and improve the edge segmentation accuracy of the original network when facing large-scale images under industrial cameras. In the experiments, the improved network in this paper has achieved 88.19% mIOU on the data set, and the segmentation accuracy of PCB components is significantly better than that of traditional FCN. Compared with the original Deeplab v3+, it has increased by 6%, and can be applied to the segmentation and recognition of components in the PCB assembly process.

關鍵字

Deeplab v3+ ； PCB Assembly ； Attention

參考文獻

F Wu, Zhang X , Kuan Y , et al. An AOI algorithm for PCB based on feature extraction[C]// World Congress on Intelligent Control & Automation. IEEE, 2008.

Lin L Z, Zhou L S , Wan J D , et al. Study of PCB Automatic Optical Inspection System Based on Mathematical Morphology[M]. IEEE, 2009.

Qiao N, Ping S. Study of improved Otsu algorithm and its ration evaluation analysis for PCB photoelectric image segmentation[J]. Optik - International Journal for Light and Electron Optics, 2014, 125(17):4784–4787.

Lin Y L, Chiang Y M, Hsu H C. Capacitor Detection in PCB Using YOLO Algorithm[C]. 2018 International Conference on System Science and Engineering. IEEE, 2018: 1-4.

Hu B, Wang J. Detection of PCB Surface Defects with Improved Faster-RCNN and Feature Pyramid Network[J]. IEEE Access, 2020, PP (99):1-1.