臺灣博碩士論文加值系統

本論文提出Tiny Yolo V2最後一層Region Layer之硬體架構設計與實現，利用Softmax與Bounding Box運算求得物件分類與位置，包含兩個主要步驟: Sigmoid運算單元及Softmax與Bounding Box運算單元的計算。首先，本層的輸入是前一層Layer 14的Output Feature Map中得到的資訊，分為座標資訊(Coordinate)、信心指數(Confidence Score)及分類資訊(Class)，先利用信心指數經由Sigmoid運算單元的計算，設定閥值以過濾沒有包含物件的分類及座標資訊，減少計算量，之後分類資訊與座標資訊再經由Softmax及Bounding Box運算單元，再次設定閥值過濾，並求得準確的物件分類與位置。此外，本論文在運算過程中會遇到Exponential及Natural Log的計算，因此在硬體架構的設計中，為使計算方便，利用座標旋轉數字計算方法(Cordic)降低運算複雜度，加速了本層的硬體運算速度。

This thesis proposed the hardware architecture design and implementation for region layer of Tiny Yolo V2 by calculating Softmax and Bounding Box to get the classification and the location of object. The proposed is composed of two steps: calculation of Sigmoid unit and Softmax unit, Bounding Box unit. First, input of region layer is the data from the Output Feature Map of layer 14, including Coordinate, Confidence Score and Class data. We use Confidence Score as the input of Sigmoid unit and set the threshold to filter the data without the object in order to reduce the calculation. Then, using Class and Coordinate as the input of Softmax unit and Bounding Box unit calculates the accurate classification and location of object. Furthermore, the process needs Exponential and Natural Log when calculating so we use Coordinate Rotation Digital Computer (Cordic) to reduce the complexity of operation and speed up the operation time in our design.

摘要 i
Abstract ii
目錄 iii
圖目錄 v
表目錄 vii
第一章 緒論 1
1.1研究動機 1
1.2研究目的 2
1.3論文架構 2
第二章 文獻探討 3
2.1 Tiny-Yolo V2 4
2.1.1分層處理 5
2.1.2 Region Layer設計概念 6
2.2分類函數 8
2.2.1 Sigmoid函數[25] 8
2.2.2 Softmax函數[25]-[26] 9
2.3 Bounding Box Regression[27]-[28] 11
2.4座標旋轉數字計算方法(Coordinate Rotation Digital Computer , Cordic) 12
2.4.1 Cordic旋轉模式(Rotation Mode) 13
2.4.2 Cordic向量模式(Vectoring Mode) 13
2.5非極大值抑制(Non-Maximum Suppression , NMS) [34] 14
第三章 研究方法 15
3.1演算法架構 15
3.2 Region Layer介紹 15
3.2.1 Sigmoid運算單元 16
3.2.2 Softmax運算單元 17
3.2.3 Bounding Box運算單元 18
3.3資料提取 21
3.4硬體架構 22
3.4.1 SRAM0-8的記憶體配置 23
3.4.2硬體單元設計 24
第四章 實驗結果與討論 32
4.1軟體模擬 32
4.2環境與效能分析 32
第五章 結論與未來工作 35
5.1結論 35
5.2未來工作 35
參考文獻 36

[1] Y. LeCun, P. Haffner, L. Bottou, Y. Bengio, "Object Recognition with Gradient-
Based Learning," Feature Grouping, 1999.
[2] Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, L. D. Jackel, “Backpropagation Applied to Handwritten Zip Code Recognition,” Neural Computation, Vol. 1, no. 4, pp. 541-551, 1989.
[3] A. Mohamed, G.E. Dahl, and G. Hinton, “Acoustic modeling using deep belief networks,” Audio, Speech, and Language Processing, IEEE Transactions on, vol. 20, no. 1, pp. 14-22, Jan. 2012.
[4] Ossama Abdel-Hamid, Abdel-rahman Mohamed, H. Jiang, L. Deng, G. Penn, D. Yu, “Convolutional Neural Networks for Speech Recognition,” IEEE/ACM Transactions on Audio, Speech, and Language Processing, Vol. 22, no. 10, Oct. 2014.
[5] F. Schroff, D. Kalenichenko, and J. Philbin, ‘‘FaceNet: A unified embedding for face recognition and clustering,’’ in Proc. IEEE Comput. Soc. Conf. Comput. Vis. Pattern Recognit, pp. 815-823, Jun. 2015.
[6] Jiang, H. and Learned-Miller, E., “Face Detection with the Faster R-Cnn,” 2017 12th IEEE International Conference on Automatic Face & Gesture Recognition (FG 2017). June. 2017.
[7] P. Sermanet, K. Kavukcuoglu, S. Chintala, and Y. LeCun, “Pedestrian detection with unsupervised multi-stage feature learning,” in CVPR, 2013.
[8] W. Ouyang, X. Wang, “Joint Deep Learning for Pedestrian Detection,” The IEEE International Conference on Computer Vision (ICCV), pp. 2056-2063, 2013.
[9] J. Li, X. Liang, S. Shen, T. Xu, J. Feng, S. Yan, “Scale-Aware Fast R-CNN for Pedestrian Detection,” IEEE Transactions on Multimedia, Vol. 20, no. 4, Apr. 2018.
[10] Shin, H.C., Roth, H.R., Gao, M., et al., “Deep convolutional neural networks for computer-aided detection: CNN architectures, dataset characteristics and transfer learning,” IEEE Trans. Med. Imaging, 35(5), pp. 1285-1298, 2016.
[11] H.C. Shin, H.R. Roth, M. Gao, L. Lu, Z. Xu, I. Nogues, J. Yao, D. Mollura, R.M. Summers, “Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures, Dataset Characteristics and Transfer Learning,” IEEE Transactions on Medical Imaging, Vol. 35, May. 2016.
[12] M. Anthimopoulos, S. Christodoulidis, L. Ebner, A. Christe, and S. Mougiakakou, ‘‘Lung pattern classification for interstitial lung diseases using a deep convolutional neural network,’’ IEEE Trans. Med. Imag., vol. 35, no. 5, pp. 1207-1216, May. 2016.
[13] K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” CoRR, vol. abs/1409.1556, 2014.
[14] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. E. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabinovich, “Going deeper with convolutions,” CoRR, vol. abs/1409.4842, 2014.
[15] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, “You Only Look Once: Unified, Real-time Object Detection,” In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 779-788, 2016.
[16] J. Redmon, A. Farhadi, “YOLO9000: Better Faster Stronger,” Computer Vision and Pattern Recognition (CVPR), 2016.
[17] J. R. Uijlings, K. E. van de Sande, T. Gevers, and A. W. Smeulders, “Selective search for object recognition,” International Journal of Computer Vision, vol. 104, no. 2, pp. 154-171, 2013.
[18] P.F. Felzenszwalb, R.B. Girshick, D. McAllester, D. Ramanan, “Object Detection with Discriminatively Trained Part-Based Models,” IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 1627-1645, 2010.
[19] P. Dollár, R. Appel, S. Belongie, and P. Perona, “Fast Feature Pyramids for Object Detection,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 36, no. 8, pp. 1532-1545, July. 2014.
[20] R. Girshick, J. Donahue, T. Darrell, and J. Malik., “Region-Based Convolutional Networks for Accurate Object Detection and Segmentation.” IEEE Transactions on Pattern Analysis and Machine Intelligence, pp. 142-158, 2016.
[21] V. Nair and G. E. Hinton, “Rectified linear units improve restricted boltzmann machines”, in ICML, pp. 807-814, 2010.
[22] Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton, “Imagenet
classification with deep convolutional neural networks,” Advances in neural
information processing systems, 2012.
[23] Iliev, Anton and Kyurkchiev, Nikolay and Markov, Svetoslav, “A Note on the New Activation Function of Gompertz Type,” vol.4, pp. 1-20, Oct. 2017.
[24] R. Girshick, “Fast r-cnn,” Proceedings of the IEEE International Conference on Computer Vision, pp. 1440-1448, 2015.
[25] I. Goodfellow, Y. Bengio, and A. Courville. “Deep learning,” Cambridge, Massachusetts: MIT Press, 2016.
[26] B. Yuan, “Efficient hardware architecture of softmax layer in deep neural network,” 29th IEEE International System-on-Chip Conference (SOCC), pp. 323-326, 2016.
[27] P. Felzenszwalb, R. Girshick, D. McAllester, and D. Ramanan, “Object detection with discriminatively trained part based models”, TPAMI, 2010.
[28] P. Sermanet, D. Eigen, X. Zhang, M. Mathieu, R. Fergus, and Y. LeCun, “OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks,” in ICLR, 2014.
[29] S. Ren, K. He, R. Girshick, J. Sun, “Faster r-cnn: Towards realtime object detection with region proposal networks,” Advances in neural information processing systems, pp. 91-99, 2015.
[30] J. E. Volder, “The CORDIC trigonometric computing technique,” IRE Transactions on Electronic Computers, vol. EC-8, pp. 330-334, Sept. 1959.
[31] Xia, Dunzhu, yu, Cheng, Wang, Yuliang, “A Digitalized Silicon Microgyroscope Based on Embedded FPGA,” Sensors (Basel, Switzerland), vol. 12, pp. 13150-13166, Dec. 2012.
[32] Rajkumar Tomar, Praveen Kumar Singh, Krishna Raj, “A Review of CORDIC Algorithms and Architectures with Applications for Efficient Designing,” International Journal of Scientific and Engineering Research (IJSER), 2013.
[33] P.K. Meher, S.Y. Park, “CORDIC designs for fixed angle of rotation,” IEEE Transaction on Very Large Scale Integration (VLSI) Systems, vol. 21, no. 2, pp. 217-228, 2013.
[34] A. Neubeck and L. Van Gool, “Efficient non-maximum suppression,” in Pattern Recognition, ICPR 2006. 18th International Conference on IEEE, vol. 3, pp. 850-855, 2006.
[35] CSDN.(2016).NMS(non maximum suppression).
Retrieved from https://blog.csdn.net/shuzfan/article/details/52711706 (June 15, 2019)
[36] Borui Jiang, Ruixuan Luo, Jiayuan Mao, Tete Xiao, Yuning Jiang, “Acquisition of Localization Confidence for Accurate Object Detection,” European Conference on Computer Vision (ECCV), 2018.
[37] Poernomo, Alvin , Kang, Dae-Ki, “Content-Aware Convolutional Neural Network for Object Recognition Task,” International journal of advanced smart convergence, vol.5, pp. 1-7, Sept. 2016.
[38] J. Sujitha, V. Ramohan Reddy, “Implementation of Log and Exponential Function in FPGA,” International Journal of Engineering Research & Technology (IJERT), vol. 3, pp. 1404-1407, Nov. 2014.
[39] Wang, Meiqi and Lu, Siyuan and Zhu, Danyang and Lin, Jun and Wang, Zhongfeng, “A High-Speed and Low-Complexity Architecture for Softmax Function in Deep Learning,” 2018 IEEE Asia Pacific Conference on Circuits and Systems, pp. 223-226,Oct. 2018.
[40] Z. Li, H. Li, X. Jiang, B. Chen, Y. Zhang, G. Du, “Efficient FPGA Implementation of Softmax Function for DNN Applications,” 2018 12th IEEE International Conference on Anti-counterfeiting, Security, and Identification (ASID), Nov. 2018.
[41] Q. Sun, Z. Di, Z. Lv, F. Song, Q. Xiang, Q. Feng, Y. Fan, X. Yu, W. Wang, “A High Speed SoftMax VLSI Architecture Based on Basic-Split,” 2018 14th IEEE International Conference on Solid-State and Integrated Circuit Technology (ICSICT), Nov. 2018.