本論文目的為透過基於深度學習之多目標檢測模型預測印刷電路板瑕疵目標，其中涵蓋模型的實現及子網路預測架構之改良。並提出混合子網路將整體效能更全面提升，與標準子網路相比擁有更精確的預測瑕疵位置能力。Soft Anchor-Point Detector(SAPD)採用anchor-point方式能將所有瑕疵類別及位置濃縮為同個位置上，較採用Anchor-based架構可降低更多訓練時間成本且擁有不遜於此架構之檢測能力。學習率決定了模型收斂速度及成效，而我們採用LR range test得出損失值與學習率關係圖，並將其轉變為變化圖來給與模型最佳的學習率。影響收斂速度還有損失函數的採用，在預選框位置的收斂方式使用Intersect over Union(IoU)將預測結果與標記進行重疊度的計算，更進一步加入對兩者方框長寬比及中心點之獎懲機制來提升原有重疊度收斂速度外更使模型擁有極佳的表現。且訓練過程中，本論文模型架構擁有較低的訓練成本外，透過優化器及搭配調節學習率僅需較少訓練週期亦可使模型擁有優異的辨識能力。採用PCB Defect及Deep PCB提供之公開印刷電路瑕疵資料集進行訓練及分析，在前者 透過混合子網路之改良及離線型融合資料強化下，mean Average Precision(mAP)在0.5至0.95重疊度下可達77.4%並能以每秒檢測影像張數(FPS)為20下執行檢測任務，後者AP可達78.2%。

In this thesis, we present a detect detection model based on deep learning to predict the detects on the printed circuit board, which covers the realization of the model and the improvement of the subnetwork. It is proposed that the Mix Subnetwork will improve the performance more comprehensively, and has a more accurate ability to predict the location of defects compared with the standard Subnetwork. Soft Anchor-Point Detector (SAPD) uses anchor-point to condense all defect categories and locations into the same location. Compared with the Anchor-based architecture, it can reduce the training time cost and reach same performance of object detection. The learning rate determines the speed of convergence and effectiveness of the model, and we use the LR range test to obtain a chart of the loss and the learning rate, and convert it into a change chart to find the model’s learning rate. The convergence of the regression bounding box we use Intersect over Union (IoU) to calculate the overlap between the prediction result and the ground truth, and obtains incentive system for the aspect ratio of the two boxes and their center points. PCB Defect and Deep PCB, which are chosen to train our model, purposed an open printed circuit board’s defects dataset. Using Mix Subnetwork with offline mix-up data augmentation method, our experiment results not only can reach 77.4% mean Average Precision(mAP) which is under challenge of 0.5 to 0.95, but also perform 20 FPS. On Deep PCB, the result we reach 78.2% mAP.

[1] W. B. Huang and P. Wei, “A PCB dataset for defects detection and classification,” IEEE Conference on Computer Vision and Pattern Recognition, vol. 14, no. 8, pp. 1-9, 2018.
[2] S. Tang, Fan He, X. Huang and J. Yang, “Online PCB Defect Detector on A New PCB Defect Dataset,” arXiv preprint arXiv: 1902.06197, 2019.
[3] K. He, X. Zhang, S. Ren and J. Sun, “Deep Residual Learning for Image Recognition,” IEEE Conference on Computer Vision and Pattern Recognition, vol. 1, pp. 770-778, 2016, 10.1109/CVPR.2016.90.
[4] S. Q. Fan, “PCB defect detection based on image processing and FPN model,” National Taiwan University of Science and Technology, Taipei, Taiwan, 2020, https://hdl.handle.net/11296/8zgydg.
[5] R. Girshick, J. Donahue, T. Darrell and J. Malik, “Rich feature hierarchies for accurate object detection and semantic segmentation,” arXiv preprint arXiv: 1311.2524, 2013.
[6] T. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan and S. Belongie, “Feature Pyramid Networks for Object Detection,” IEEE Conference on Computer Vision and Pattern Recognition, pp. 936-944, 2017, 10.1109/CVPR.2017.106.
[7] C. Zhu, F. Chen, Z. Shen and M. Savvides, “Soft Anchor-Point Object Detection,” arXiv preprint arXiv: 1911.12448, 2020.
[8] S. Ren, K. He, R. Girshick and J. Sun, "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 6, pp. 1137-1149, 2017, 10.1109/TPAMI.2016.2577031.
[9] T. Lin, P. Goyal, R. Girshick, K. He and P. Dollár, "Focal Loss for Dense Object Detection," 2017 IEEE International Conference on Computer Vision, pp. 2999-3007, 2017, 10.1109/ICCV.2017.324.
[10] Z. Tian, C. Shen, H. Chen and T. He, "FCOS: Fully Convolutional One-Stage Object Detection," 2019 IEEE/CVF International Conference on Computer Vision (ICCV), pp. 9626-9635, 2019, 10.1109/ICCV.2019.00972.
[11] C. Zhu, Y. He and M. Savvides, "Feature Selective Anchor-Free Module for SingleShot Object Detection," 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 840-849, 2019, 10.1109/CVPR.2019.00093.
[12] S. Qiao, H. Wang, C. Liu, W. Shen and A. Yuille, “Micro-Batch Training with BatchChannel Normalization and Weight Standardization,” arXiv preprint arXiv: 1903.10520, 2020.
[13] Y. Wu and K. He, “Group Normalization,” Proceedings of the European Conference on Computer Vision (ECCV), pp. 3-19, 2018.
[14] Z. Zheng, P. Wang, W. Liu, J. Li, R. Ye and D. Ren, “Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression,” arXiv preprint arXiv: 1911.08287, 2019.
[15] L. N. Smith, “Cyclical Learning Rates for Training Neural Networks,” 2017 IEEE Winter Conference on Applications of Computer Vision, pp. 464-472, 2017, 10.1109/WACV.2017.58.
[16] A. Elhassouny and F. Smarandache, “Trends in deep convolutional neural Networks architectures: a review,” International Conference of Computer Science and Renewable Energies, pp. 1-8, 2019, 10.1109/ICCSRE.2019.8807741.
[17] K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” arXiv preprint arXiv:1409.1556, 2014.
[18] A. Krizhevsky, I. Sutskever, and G. E. Hinton, “Imagenet classification with deep convolutional neural networks,” Advances in neural information processing systems, pp. 1097-1105, 2012.
[19] M. D. Zeiler and R. Fergus, “Visualizing and understanding convolutional networks,” European conference on computer vision, pp. 818-833, 2014.
[20] V. Thakkar, S. Tewary and C. Chakraborty, “Batch Normalization in Convolutional Neural Networks - A comparative study with CIFAR-10 data,” 2018 Fifth International Conference on Emerging Applications of Information Technology (EAIT), pp. 1-5, 2018, 10.1109/EAIT.2018.8470438.
[21] V. Nair and G. E. Hinton, “Rectified Linear Units Improve Restricted Boltzmann Machines,” ICML, 2010.
[22] K. He, R. Girshick and P. Dollar, "Rethinking ImageNet Pre-Training," 2019 IEEE/CVF International Conference on Computer Vision, pp. 4917-4926, 2019, 10.1109/ICCV.2019.00502.
[23] A. Kolesnikov, L. Beyer, X. Zhai, J. Puigcerver, J. Yung, S. Gelly and N. Houlsby, “Big Transfer (BiT): General Visual Representation Learning ,” arXiv preprint arXiv: 1912.11370, 2020.
[24] K. Chen, Y. Cao, C. C. Loy, D. Lin, C. Feichtenhofer, “Feature Pyramid Grids,” arXiv preprint arXiv: 2004.03580, 2020.
[25] Z. Wang, C. Wang, Z. Su and J. Chen, “Dense Feature Pyramid Grids Network for Single Image Deraining,” ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp. 2025-2029, 2021, 10.1109/ICASSP39728.2021.9415034.
[26] J. Redmon, S. Divvala, R. Girshick and A. Farhadi, “You Only Look Once: Unified, Real-Time Object Detection,” arXiv preprint arXiv: 1506.02640, 2015.
[27] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C. Y. Fu and A. C. Berg, “SSD: Single Shot MultiBox Detector,” arXiv preprint arXiv:1512.02325, 2016.
[28] R. Girshick, “Fast R-CNN,” 2015 IEEE International Conference on Computer Vision (ICCV), pp. 1440-1448, 2015, 10.1109/ICCV.2015.169.
[29] S. Ren, K. He, R. Girshick and J. Sun, “Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 6, pp. 1137-1149, 2017, 10.1109/TPAMI.2016.2577031.
[30] K. He, G. Gkioxari, P. Dollár and R. Girshick, “Mask R-CNN,” 2017 IEEE International Conference on Computer Vision (ICCV), pp. 2980-2988, 2017, 10.1109/ICCV.2017.322.
[31] J. R. R. Uijlings, K. E. A. van de Sande, T. Gevers and A. W. M. Smeulders, “Selective Search for Object Recognition,” International Journal of Computer Vision, vol. 104, pp. 154-171, 2013.
[32] J. R. R. Uijlings, K. E. A. van de Sande, T. Gevers and A. W. M. Smeulders, “The PASCAL Visual Object Classes (VOC) Challenge,” International Journal of Computer Vision, vol. 88, pp. 303-338, 2010.
[33] T. Lin, M. Maire, S. Belongie, L. Bourdev, R. Girshick, J. Hays, P. Perona, D. Ramanan, C. L. Zitnick and P. Dollár, “Microsoft COCO: Common Objects in Context,” arXiv preprint arXiv:1405.0312, 2015.
[34] J. Deng, W. Dong, R. Socher, L. Li, Kai Li and Li Fei-Fei, “ImageNet: A large-scale hierarchical image database,” 2009 IEEE Conference on Computer Vision and Pattern Recognition, pp. 248-255, 2009, 10.1109/CVPR.2009.5206848.
[35] Y. Wu et al., “Rethinking Classification and Localization for Object Detection,” 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pp. 10183-10192, 2020, 10.1109/CVPR42600.2020.01020.
[36] H. Rezatofighi, N. Tsoi, J. Gwak, A. Sadeghian, I. Reid and S. Savarese, “Generalized Intersection Over Union: A Metric and a Loss for Bounding Box Regression,” IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 658-666, 2019, 10.1109/CVPR.2019.00075.
[37] L. Loshchilov and F. Hutter, “SGDR: Stochastic Gradient Descent with Warm Restarts,” arXiv preprint arXiv:1608.03983, 2016.
[38] D. P. Kingma and J. Ba, “Adam: A Method for Stochastic Optimization,” arXiv preprint arXiv:1412.6980, 2014.
[39] S. Zhang, L. Wen, X. Bian, Z. Lei and S. Z. Li, “Single-Shot Refinement Neural Network for Object Detection,” 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 4203-4212, 2018, 10.1109/CVPR.2018.00442.
[40] G. Jin, R. Taniguchi and F. Qu, “Auxiliary Detection Head for One-Stage Object Detection,” IEEE Access, vol. 8, pp. 85740-85749, 2020, 10.1109/ACCESS.2020.2992532.
[41] T. Kong, F. Sun, H. Liu, Y. Jiang, L. Li and J. Shi, “FoveaBox: Beyound Anchor-Based Object Detection,” IEEE Transactions on Image Processing, vol. 29, pp. 7389-7398, 2020, 10.1109/TIP.2020.3002345.
[42] J. Hosang, R. Benenson, and B. Schiele “A Convnet for Non-maximum Suppression,” arXiv preprint arXiv:1511.06437, 2016.
[43] S. Liu, D. Huang and Y. Wang, “Adaptive NMS: Refining Pedestrian Detection in a Crowd,” arXiv preprint arXiv:1904.03629, 2019.
[44] A. Shepley, G. Falzon and P. Kwan, “Confluence: A Robust Non-IoU Alternative to Non-Maxima Suppression in Object Detection,” arXiv preprint arXiv:2012.00257, 2020.
[45] Z. Cai and N. Vasconcelos, “Cascade R-CNN: Delving Into High Quality Object Detection,” 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 6154-6162, 2018, 10.1109/CVPR.2018.00644.
[46] H. Zhang, H. Chang, B. Ma, S. Shan and X. Chen, “Cascade RetinaNet: Maintaining Consistency for Single-Stage Object Detection,” arXiv preprint arXiv:1907.06881, 2019.
[47] Q. Lin, Y. Ding, H. Xu, W. Lin, J. Li and X. Xie, “ECascade-RCNN: Enhanced Cascade RCNN for Multi-scale Object Detection in UAV Images,” 2021 7th International Conference on Automation, Robotics and Applications (ICARA), pp. 268-272, 2021, 10.1109/ICARA51699.2021.9376456.
[48] I. Loshchilov and F. Hutter “Decoupled Weight Decay Regularization,” arXiv preprint arXiv:1711.05101, 2019