數十年來，深度學習已經得到越來越多的發展和大規模投資。基於物體檢測的方法已被證明在各種領域真正有效，如高級駕駛員輔助系統，視覺引導機器人手臂控制，自動光學檢查，人臉識別和運動分析。 SSD [1]，RetinaNet [2]和YOLO [3]已經證明了其功能，並且由於在最小化性能和速度之間的權衡方面取得了顯著的改進，因此在單級物體檢測方法中引起了相當大的關注。然而，在復雜場景中檢測和分類細粒度或重疊對象仍然是當前基於一階段對象檢測的深度學習方法的一大弱點。
球柵陣列（BGA）芯片生產是半導體行業的先鋒之一。芯片性能測試需要精確的自動化流程，為將深度學習方法應用於集成電路工業打開了大門。對於BGA芯片製造而言，由於缺陷，將焊球粘附到芯片表面貼裝封裝上的過程仍然是製造商的挑戰。
本文的主要貢獻在於：（1）應用現有的深度學習方法解決BGA芯片上微小缺陷的檢測和分類問題; （2）與現有技術方法相比，通過增強YOLOv3算法的弱點，建立改進的目標檢測算法，但仍保持基本特徵和精度與速度之間的良好平衡; （3）將該修改算法應用於與球柵陣列芯片缺陷檢測和分類相關的實際項目中。

Deep learning [8] has been increasingly developed and massively invested for several decades. Object Detection-based approaches have proved genuinely effective in various areas such as Advanced Driver-assistance System, Visual Guided Robot Arm Control, Automatic Optical Inspection, Face Recognition and Sports Analysis. SSD [11], RetinaNet [12] and YOLO [15], [17] have proven capability and attracted considerable attention for One-Stage Object Detection methods due to the significant improvement in minimizing the trade-off between performance and speed. However, detecting and classifying fine-grained or overlapping objects in complex scenes have been being still big weaknesses of current One-stage Object Detection-based Deep Learning methods.
Ball-Grid-Array (BGA) Chip production is one of the spearheads in Semiconductor Industry. The Chip performance testing requires precise and automated processes that opens the door for applying Deep Learning methods into Integrated Circuit Industries. For the BGA Chip manufacture, the process of sticking solder balls onto the chip surface-mount packages has still been challenging manufacturers because of the Defects.
The main contributions of this thesis are emphasized on: (1) Apply current Deep Learning methods to solving the problem of Detecting and Classifying tiny defects on BGA Chips; (2) building an improved Object Detection algorithm by enhancing the weaknesses of YOLOv3 algorithm compared to state-of-the-art methods but still keeping the basic characteristics and the good trade-off between accuracy and speed; (3) implementing this modified algorithm into practical projects related to Detecting and Classifying Defects in Ball-Grid-Array Chips.

[1] J. Deng, W. Dong, R. Socher, L. Li, K. Li and F. Li, “ImageNet: A large-scale hierarchical image database,” The IEEE Conference on Computer Vision and Pattern Recognition, pp. 248-255, 2009.
[2] R. Girshick, J. Donahue, T. Darrell and J. Malik, "Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation," Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 580-587, 2014.
[3] R. Girshick, "Fast R-CNN," Proceedings of the IEEE International Conference on Computer Vision, pp. 1440-1448, 2015.
[4] G. Huang, Z. Liu, L. V. D. Maaten and K. Q. Weinberger, “Densely Connected Convolutional Networks”, arXiv preprint arXiv:1608.06993, 2018.
[5] K. He, X. Zhang, S. Ren, J. Sun, “Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 37, No. 9, pp. 1904-1916, 2015.
[6] K. He, "Deep Residual Learning for Image Recognition," Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 770-778, 2016.
[7] J. Huang et al., "Speed/accuracy trade-offs for modern convolutional object detectors," Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 3296-3297, 2017.
[8] Y. LeCun, Y. Bengio, and G. Hinton, "Deep learning", Nature, vol. 521, pp. 436-444, 2015.
[9] T. Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Dollar, and C. L. Zitnick, “Microsoft COCO: Common objects in context,” Proceedings of European Conference on Computer Vision, pp. 740-755, 2014.
[10] J. Long, E. Shelhamer, T. Darrell, “Fully Convolutional Networks for Semantic Segmentation,” The IEEE Conference on Computer Vision and Pattern Recognition, pp. 3431-3440, 2015.
[11] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, and S. Reed, “SSD: Single shot multibox detector,” Proceedings of European Conference on Computer Vision, pp. 21-37, 2016.
[12] T. Lin, P. Goyal, R. Girshick, K. He, P. Dollar, “Focal Loss for Dense Object Detection,” Proceedings of International Conference on Computer Vision, pp. 2980-2988, 2017.
[13] X. Li, J. Wu, Z. Lin, H. Liu and H. Zha, “Recurrent Squeeze-and-Excitation Context Aggregation Net for Single Image Deraining”, ECCV, 2018.
[14] S. Ren, K. He, R. Girshick, and J. Sun, "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks," Advances in Neural Information Processing Systems, pp. 91-99, 2015.
[15] J. Redmon, A. Farhad, “YOLO9000: Better, Faster, Stronger,” arXiv preprint arXiv:1612.08242, 2016.
[16] J. Redmon, S. Divvala, R. Girshick and A. Farhadi, “You only look once: Unified, real-time object detection,” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 779-788, 2016.
[17] J. Redmon, A. Farhadi, “YOLOv3: An Incremental Improvement,” arXiv preprint arXiv:1804.02767, 2018.
[18] W. Shi, J. Caballero, F. Huszár, J. Totz, A. P. Aiken, R. Bishop, D. Rueckert and Z. Wang, “Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network”, CVPR, 2016.
[19] Z. Shen, Z. Liu, J. Li, Y. Chen and X. Xue, “DSOD: Learning Deeply Supervised Object Detectors from Scratch”, ICCV, 2017. s