現代電子產品都朝向輕薄短小與高度整合化發展，柔性印刷電路板(Flexible Printed Circuit Board, FPCB)於製程中更易產生微小且難以辨識之缺陷，對自動光學 檢測系統(Automated Optical Inspection, AOI)構成極大挑戰。為因應此問題，本研究 目的是提出一種結合注意力機制與通道剪枝策略的輕量化物件偵測模型(LiteC-GAM YOLO)，專為FPCB多類別缺陷辨識任務而設計。
Lite C-GAM YOLO模型在原有 YOLOv8s 架構中導入 GAM(Global Attention Module(以強化關鍵特徵的提取能力，並透過1×1卷積進行特徵通道壓縮，以濃縮 資訊。進一步結合通道剪枝技術，能有效移除冗餘通道，不僅降低模型參數與運算 量，更顯著減少整體計算成本，同時維持高準確率。實驗結果顯示，在涵蓋九種類 別缺陷的缺陷資料集上，尤其對「Pin-hole」與「Copper-residue」等難以辨識之缺陷 具備更高的辨識能力，皆優於原始YOLOv8s與其他YOLO系列模型的辨識效能，展 現其卓越的檢測能力。
綜合上述成果，LiteC-GAMYOLO模型兼具高準確率、低資源消耗與良好泛 化能力，特別適用於實際製造環境中的缺陷檢測任務。相較於原始YOLOv8s模型， 本模型在導入剪枝技術後，mAP提升3.9%，參數量減少達45%，成功實現高精度與 輕量化的雙重目標。此設計不僅有效降低因人工疲勞造成的漏檢與誤判風險，有助 於防止缺陷品流出，進一步提升整體生產效率，展現出在智慧製造領域中的高度應 用潛力。

With the ongoing trend toward miniaturization and high integration in modern electronic products, Flexible Printed Circuit Boards (FPCBs) are increasingly prone to producing subtle and difficult-to-detect defects during the manufacturing process. These defects pose significant challenges to Automated Optical Inspection (AOI) systems. To address this issue, this study proposes a lightweight object detection model, named Lite C-GAM YOLO, which integrates attention mechanisms and channel pruning strategies, specifically designed for multi-class defect detection in FPCBs.
The Lite C-GAM YOLO model incorporates the Global Attention Module (GAM) into the original YOLOv8 architecture to enhance the extraction of critical features. Additionally, a 1×1 convolution is employed to compress feature channels, thereby condensing informative representations. By further integrating channel pruning techniques, the model effectively removes redundant channels, which not only reduces the number of parameters and computational load but also significantly lowers the overall computational cost while maintaining high detection accuracy. Experimental results on a dataset containing nine types of FPCB defects demonstrate that the proposed model exhibits superior recognition performance, especially for hard-to-detect defects such as “Pin-hole” and “Copper-residue,” outperforming the original YOLOv8 and other YOLO variants.
In summary, the Lite C-GAM YOLO model achieves a balanced trade-off between high accuracy, low resource consumption, and strong generalization capability, making it particularly suitable for real-world defect detection applications in manufacturing environments. Compared to the original YOLOv8, the proposed model achieves a 3.9% improvement in mAP after pruning and reduces the parameter count by up to 45%, successfully fulfilling the goals of high precision and model compactness. This design not only reduces the risk of missed or incorrect detections caused by operator fatigue but also helps prevent defective products from reaching the market, ultimately improving production efficiency and demonstrating high potential for practical deployment in intelligent manufacturing.

[1] Yasuhiko Hara, Nobuyuki Akiyama, and Koichi Karasaki, "Automatic Inspection System for Printed Circuit Boards," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. PAMI-5, no. 6, pp. 623–630, Nov. 1983.
[2] Roger Z. Liu, Yun Qing Shi, William F. Kosonocky, and F. P. Wiggins, "Infrared Solder Joint Inspection on Surface Mount Printed Circuit Boards," in 1995 38th Midwest Symposium on Circuits and Systems (MWSCAS), Rio de Janeiro, Brazil, pp. 145–148, Aug. 1995.
[3]Yohei Takada, Tokiko Shiina, Hiroyasu Usami, Yuji Iwahori, and M. K. Bhuyan, "Defect Detection and Classification of Electronic Circuit Boards Using Keypoint Extraction and CNN Features," in 2017 Ninth International Conference on Pervasive Patterns and Applications (PATTERNS 2017), Nice, France, pp. 113–116, Jul. 2017.
[4] Manjun Tian, Xiali Li, Shihan Kong, Licheng Wu, and Junzhi Yu, "Pruning-Based YOLOv4 Algorithm for Underwater Gabage Detection," in 2021 40th Chinese Control Conference (CCC), Shanghai, China, pp. 4008–4013, Jul. 2021.
[5] Ross Girshick, Jeff Donahue, Trevor Darrell, and Jitendra Malik, "Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation," in 2014 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Columbus, OH, USA, pp. 580–587, Jun. 2014.
[6] Ross Girshick, "Fast R-CNN," in 2015 IEEE International Conference on Computer Vision (ICCV), Santiago, Chile, pp. 1440–1448, Dec. 2015.
[7] Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun, "Faster R-CNN: Towards real-time object detection with region proposal networks," in Advances in Neural Information Processing Systems, vol. 28, Montreal, Canada, pp. 91–99, Dec. 2015.
[8] Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi, "You Only Look Once: Unified, Real-Time Object Detection," in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, pp. 779–788, Jun. 2016.
[9] Joseph Redmon and Ali Farhadi, "YOLO9000: Better, Faster, Stronger," in 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, pp. 7263–7271, Jul. 2017.
[10] Joseph Redmon and Ali Farhadi, "YOLOv3: An Incremental Improvement ," arXiv preprint arXiv:1804.02767, Apr. 2018.
[11] Alexey Bochkovskiy, Chien-Yao Wang, and Hong-Yuan Mark Liao, "YOLOv4: Optimal Speed and Accuracy of Object Detection," arXiv preprint arXiv:2004.10934, Apr. 2020.
[12] Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng Yang Fu, and Alexander C. Berg, "SSD: Single Shot MultiBox Detector," in 2016 European Conference on Computer Vision (ECCV), Amsterdam, The Netherlands, pp. 21–37, Oct. 2016.
[13] Ultralytics, "YOLOv5," 2020. [Online]. Available: https://github.com/ultralytics/yolov5.
[14] Chuyi Li, Lulu Li, Hongliang Jiang, Kaiheng Weng, Yifei Geng, Liang Li, Zaidan Ke, Qingyuan Li, Meng Cheng, Weiqiang Nie, Yiduo Li, Bo Zhang, Yufei Liang, Linyuan Zhou, Xiaoming Xu, Xiangxiang Chu, Xiaoming Wei, and Xiaolin Wei, "YOLOv6: A Single-Stage Object Detection Framework for Industrial Applications," arXiv preprint arXiv:2209.02976, Sep. 2022.
[15] Chien-Yao Wang, Alexey Bochkovskiy, and Hong-Yuan Mark Liao, "YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors," arXiv preprint arXiv:2207.02696, Jul. 2022.
[16] Ultralytics, "YOLOv8," 2023. [Online]. Available: https://github.com/ultralytics/ultralytics.
[17] Ultralytics, "YOLOv9," 2024. [Online]. Available: https://github.com/ultralytics/ultralytics.
[18] Ultralytics, "YOLOv10," 2024. [Online]. Available: https://github.com/ultralytics/ultralytics.
[19] Ultralytics, "YOLOv11," 2024. [Online]. Available: https://github.com/ultralytics/ultralytics.
[20] Ultralytics, "YOLOv12," 2024. [Online]. Available: https://github.com/ultralytics/ultralytics.
[21] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E. Hinton, "ImageNet Classification with Deep Convolutional Neural Networks," in Advances in Neural Information Processing Systems, vol. 25, Lake Tahoe, NV, USA, pp. 1097–1105, Dec. 2012.
[22] Karen Simonyan and Andrew Zisserman, "Very Deep Convolutional Networks for Large-Scale Image Recognition," arXiv preprint arXiv:1409.1556, Sep. 2014.
[23] Mingxing Tan and Quoc V. Le, "EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks," in 2019 36th International Conference on Machine Learning (ICML), Long Beach, CA, USA, vol. 97, pp. 6105–6114, Jun. 2019.
[24] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin, "Attention Is All You Need," in Advances in Neural Information Processing Systems, vol. 30, Long Beach, CA, USA, pp. 5998–6008, Dec. 2017.
[25] Yichao Liu, Zongru Shao, and Nico Hoffmann, "Global Attention Mechanism: Retain Information to Enhance Channel-Spatial Interactions," arXiv preprint arXiv:2112.05561, Dec. 2021.
[26] Hao Li, Asim Kadav, Igor Durdanovic, Hanan Samet, and Hans Peter Graf, "Pruning Filters for Efficient ConvNets," in 2017 International Conference on Learning Representations (ICLR), Toulon, France, Apr. 2017.
[27] Yihui He, Xiangyu Zhang, and Jian Sun, "Channel Pruning for Accelerating Very Deep Neural Networks," in 2017 IEEE International Conference on Computer Vision (ICCV), Venice, Italy, pp. 1389-1397, Oct. 2017.
[28] Yang He, Ping Liu, Ziwei Wang, Zhilan Hu, and Yi Yang, "Filter Pruning via Geometric Median for Deep Convolutional Neural Networks Acceleration," in 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Long Beach, CA, USA, pp. 4340-4349, Jun. 2019.
[29] Shu Liu, Lu Qi, Haifang Qin, Jianping Shi, and Jiaya Jia, "Path aggregation network for instance segmentation," in 2018 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, UT, USA, pp. 8759-8768, Jun. 2018.
[30] Xiang Li, Wenhai Wang, Lijun Wu, Shuo Chen, Xiaolin Hu, Jun Li, Jinhui Tang, and Jian Yang, "Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection," in 34th Conference on Neural Information Processing Systems (NeurIPS 2020), vol. 33, pp. 21002–21012, Vancouver, Canada, Dec. 2020.
[31] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun, "Deep Residual Learning for Image Recognition," in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, pp. 770-778, Jun. 2016.
[32] En Li, Zhi Zhou, Xu Chen, Liekang Zeng, and Sheng Zhou, "Edge AI: On-Demand Accelerating Deep Neural Network Inference via Edge Computing," arXiv preprint arXiv:1910.05316, Oct. 2019.
[33] Jie Hu, Li Shen, and Gang Sun, "Squeeze-and-Excitation Networks," in 2018 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, UT, USA, pp. 7132-7141, Jun. 2018.
[34] Sanghyun Woo, Jongchan Park, Joon-Young Lee, and In So Kweon, "CBAM: Convolutional Block Attention Module," in 2018 European Conference on Computer Vision (ECCV), Munich, Germany, pp. 3-19, Sep. 2018.
[35] Andrew Howard, Mark Sandler, Grace Chu, Liang-Chieh Chen, Bo Chen, Mingxing Tan, and Quoc V. Le, "Searching for MobileNetV3," in 2019 IEEE/CVF International Conference on Computer Vision (ICCV), Seoul, South Korea, pp. 1314-1324, Oct. 2019.
[36] Baozhou Zhu, Peter Hofstee, Jinho Lee, and Zaid Al-Ars, "An Attention Module for Convolutional Neural Networks," arXiv preprint arXiv:2108.08205, Aug. 2021.
[37] Yang Lu, "Industry 4.0: A survey on technologies, applications and open research issues," Journal of Industrial Information Integration, vol. 6, pp. 1-10, Jun. 2017.
[38] Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio, "Neural Machine Translation by Jointly Learning to Align and Translate," arXiv preprint arXiv:1409.0473, May 2015.
[39] Fan Meng, Haoyu Cheng, Kai Li, Hongwei Luo, Xuefeng Guo, Guangming Lu, and Xiaolin Sun, "Pruning Filter in Filter," in 34th Conference on Neural Information Processing Systems (NeurIPS 2020), vol. 33, pp. 17629–17639, Dec. 2020.
[40] Heng Xie, Yong Li, Xiaoxue Li, and Liang He, "A Method for Surface Defect Detection of Printed Circuit Board Based on Improved YOLOv4," in 2021 IEEE Second International Conference on Big Data, Artificial Intelligence and Internet of Things Engineering (ICBAIE), Nanchang, China, pp. 851-857, May 2021.
[41] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby, "An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale" arXiv preprint arXiv:2010.11929, Oct. 2020.
[42] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo, "Swin Transformer: Hierarchical Vision Transformer using Shifted Windows," in 2021 IEEE/CVF International Conference on Computer Vision (ICCV), Montreal, QC, Canada, pp. 10012–10022, Oct. 2021.