電腦視覺處理已從傳統方法（如精選特徵）到當前最先進的捲積神經網絡（CNN）模型有了顯著改進。使用CNN進行特徵提取是一個自主過程，由機器從數據集中學習。各種因素在此特徵提取中起著主要作用，這可能會使模型容易受到對抗性攻擊和有缺陷的學習。本文解決了三個問題子集，詳細闡述了CNN的強健性和模型訓練過程。第一個問題是處理二維分形布朗運動和二維分形高斯噪聲的分類，二維分形高斯噪聲是具有各種Hessian值的粒子的隨機運動。該數據集在圖像中包含對抗功能。我們評估了各種CNN模型進行分類，並提出了用於分類的最佳集成神經網絡模型，其準確度為97.6％。第二個問題是用於自主農場管理的物件檢測，定位，跟踪和方向檢測。我們已經提出並實施了人工智能視覺監視，以監視農場中的各種物體及其行為。它還在具有基於IP的監視系統的Nvidia Jetson AGX Xavier和Nvidia RTX 2080Ti上實施，以提高自主農場管理的效率。第三個問題是處理對抗性攻擊。我們評估了COCO數據集上具有各種超參數的對抗攻擊，並推理出最佳實踐以創建可靠的CNN模型。本文詳細介紹了各種CNN模型的實現和工作原理，重點是提高所提出模型的整體強健性。同時我們還討論了對抗功能和對抗攻擊。

Computer-vision processing has been significantly improved from the traditional methods such as hand-picked features to the current state of art convolution neural network (CNN) models. Feature extraction using CNN is an autonomous process which is learned by the machine from the dataset. Various factors play a major role in this feature extraction, which can make the model susceptible to adversarial attacks and defective learning. This Thesis tackles three subsets of problems to elaborate on the robustness of CNN and model training process. The first problem is to deal with classification of 2D fractional Brownian motion and 2D fractional Gaussian noise which is a random motion of particles with various Hessian values. This Dataset contains adversarial feature in the images. This thesis has evaluated various CNN models for classification and proposed the best ensemble-CNN model for classification with accuracy of 97.6%. The second problem is to deal with object detection, localization, tracking and direction detection for autonomous farm management. The proposed artificial intelligence visual surveillance model to monitor various objects and their behaviors in the farm using Edge Computing. It is implemented on the Nvidia Jetson AGX Xavier and Nvidia RTX 2080Ti with IP-based surveillance system to improve the efficiency of autonomous farm management. The third problem is to deal with inferencing adversarial attacks. This thesis has evaluated adversarial attacks with various hyper-parameters and regularization techniques on the COCO dataset and reasoning the best practices to create a robust CNN model. In addition, adversarial features and adversarial attacks are discussed briefly. This thesis also elaborates on the implementation and the workings of various CNN models with focus on improving the overall robustness of the proposed models.

摘要...i
Abstract...iii
Acknowledgements...v
Table of Contents...vi
List of Tables...viii
List of Figures...ix
Chapter 1 Introduction...1
1.1 Review of Deep Learning...1
1.2 Convolutional Neural Networks...2
1.3 Adversarial Features...3
1.4 Adversarial Attacks...4
1.5 Motivation and Contribution...5
1.6 Organization of the Thesis...7
Chapter 2 Mathematical Background...8
2.1 Working of Neural Networks...8
2.2 Convolutional Layer...9
2.3 Pooling Layer...11
2.4 Activation Functions...11
2.5 MiniBatch-GD vs GD vs Adam...13
2.6 CNN Working...15
Chapter 3 Classification of 2D Fractional Brownian Motion and 2D Fractional Gaussian Noise...17
3.1 Introduction...17
3.2 Dataset...18
3.2.1 Two Dimensional Fractional Brownian Motion...19
3.2.2 Two Dimensional Fractional Gaussian Noise...25
3.3 Methods...30
3.4 Observation and Results...43
3.5 Concluding Remarks...52
Chapter 4 Artificial Intelligence Visual Surveillance for Autonomous Farm Management...54
4.1 Introduction...54
4.2 Object Detection Techniques...55
4.3 Mathematical Analysis of 3D convolution Filter and Back-Propagation...56
4.4 Dataset...57
4.5 Methods...60
4.5.1 Object Detection and Localization...62
4.5.2 Tracking and Direction Detection...66
4.5.3 Distance Estimation...67
4.6 Hardware and Software Implementation...68
4.7 Results...71
4.7.1 Distance Estimation...78
4.7.2 Comparative Analysis...79
4.8 Concluding Remarks...79
Chapter 5 Adversarial attacks with Various Hyper-Parameters on the COCO Dataset and Defense Techniques...81
5.1 Introduction to Adversarial Attacks...81
5.2 Adversarial Attacks...83
5.3 Tackling Adversarial Attacks...84
5.4 Experiments Results on COCO Dataset...85
5.5 Concluding Remarks...93
Chapter 6 Conclusions and Future Research...94
6.1 Conclusions...94
6.2 Future Works...94
References...96
Extended Abstract...104

[1] Gu, Jiuxiang, Zhenhua Wang, Jason Kuen, Lianyang Ma, Amir Shahroudy, Bing Shuai, Ting Liu et al. "Recent Advances in Convolutional Neural Networks." Pattern Recognition 77 (2018): 354-377.
[2] Davenport, Thomas H., and Rajeev Ronanki. "Artificial Intelligence for The Real World." Harvard Business Review 96, no. 1 (2018): 108-116.
[3] Wang, Xizhao, Yanxia Zhao, and Farhad Pourpanah. "Recent Advances in Deep Learning." (2020): 1-4.
[4] Hubel, David H., And Torsten N. Wiesel. "Receptive Fields of Single Neurons in The Cat's Striate Cortex." The Journal of Physiology 148, No. 3 (1959): 574.
[5] Fukushima, Kunihiko, And Sei Miyake. "Neocognitron: A Self-Organizing Neural Network Model for A Mechanism of Visual Pattern Recognition." In Competition and Cooperation in Neural Nets, Pp. 267-285. Springer, Berlin, Heidelberg, 1982.
[6] Le Cun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, And L. D. Jackel. "Handwritten Digit Recognition with A Back-Propagation Network." In Advances in Neural Information Processing Systems. 1990.
[7] Chen, Xue-Wen, And Xiaotong Lin. "Big Data Deep Learning: Challenges and Perspectives." IEEE Access 2 (2014): 514-525.
[8] Najafabadi, Maryam M., Flavio Villanustre, Taghi M. Khoshgoftaar, Naeem Seliya, Randall Wald, And Edin Muharemagic. "Deep Learning Applications and Challenges in Big Data Analytics." Journal of Big Data 2, No. 1 (2015): 1.
[9] Zhang, Quan-Shi, And Song-Chun Zhu. "Visual Interpretability for Deep Learning: A Survey." Frontiers of Information Technology & Electronic Engineering 19, No. 1 (2018): 27-39.
[10] Ilyas, Andrew, Shibani Santurkar, Dimitris Tsipras, Logan Engstrom, Brandon Tran, And Aleksander Madry. "Adversarial Examples Are Not Bugs, They Are Features." In Advances in Neural Information Processing Systems, Pp. 125-136. 2019.
[11] Liu, Li, Wanli Ouyang, Xiaogang Wang, Paul Fieguth, Jie Chen, Xinwang Liu, And Matti Pietikäinen. "Deep Learning for Generic Object Detection: A Survey." International Journal of Computer Vision 128, No. 2 (2020): 261-318.
[12] Yuan, Xiaoyong, Pan He, Qile Zhu, And Xiaolin Li. "Adversarial Examples: Attacks and Defenses for Deep Learning." IEEE Transactions On Neural Networks and Learning Systems 30, No. 9 (2019): 2805-2824.
[13] Gilmer, Justin, Nicolas Ford, Nicholas Carlini, And Ekin Cubuk. "Adversarial Examples are a Natural Consequence of Test Error in Noise." In International Conference On Machine Learning, Pp. 2280-2289. 2019.
[14] Akhtar, Naveed, And Ajmal Mian. "Threat of Adversarial Attacks On Deep Learning in Computer Vision: A Survey." IEEE Access 6 (2018): 14410-14430.
[15] Madry, Aleksander, Aleksandar Makelov, Ludwig Schmidt, Dimitris Tsipras, And Adrian Vladu. "Towards Deep Learning Models Resistant to Adversarial Attacks." Arxiv Preprint Arxiv:1706.06083 (2017).
[16] Barnard, Etienne. "Optimization for Training Neural Nets." IEEE Transactions on Neural Networks 3, No. 2 (1992): 232-240.
[17] Ellacott, Stephen W., John C. Mason, And Iain J. Anderson, Eds. Mathematics of Neural Networks: Models, Algorithms and Applications. Vol. 8. Springer Science & Business Media, 2012.
[18] Karlik, Bekir, And A. Vehbi Olgac. "Performance Analysis of Various Activation Functions in Generalized MLP Architectures of Neural Networks." International Journal of Artificial Intelligence and Expert Systems 1, No. 4 (2011): 111-122.
[19] Alfarra, Motasem, Slavomir Hanzely, Alyazeed Albasyoni, Bernard Ghanem, And Peter Richtarik. "Adaptive Learning of the Optimal Mini-Batch Size of SGD." Arxiv Preprint Arxiv:2005.01097 (2020).
[20] Khan, Mohammad Emtiyaz, Didrik Nielsen, Voot Tangkaratt, Wu Lin, Yarin Gal, And Akash Srivastava. "Fast and Scalable Bayesian Deep Learning by Weight-Perturbation in Adam." Arxiv Preprint Arxiv:1806.04854 (2018).
[21] Zhou, Pan, Jiashi Feng, Chao Ma, Caiming Xiong, And Steven Chu Hong Hoi. "Towards Theoretically Understanding Why SGD Generalizes Better Than ADAM in Deep Learning." Advances in Neural Information Processing Systems 33 (2020).
[22] Zhang, Zhifei. "Derivation of Backpropagation in Convolutional Neural Network (CNN)." University of Tennessee, Knoxville, TN (2016).
[23] Mizutani, Hikaru. "The Back Propagation Method for CNN." In Proceedings of IEEE International Symposium On Circuits and Systems-ISCAS'94, Vol. 6, Pp. 463-466. IEEE, 1994.
[24] Heneghan, C., S. B. Lowen, And Malvin C. Teich. "Two-Dimensional Fractional Brownian Motion: Wavelet Analysis and Synthesis." In Proceeding of Southwest Symposium On Image Analysis and Interpretation, Pp. 213-217. IEEE, 1996.
[25] Reed, Irving S., P. C. Lee, And Trieu-Kien Truong. "Spectral Representation of Fractional Brownian Motion in N Dimensions and Its Properties." IEEE Transactions On Information Theory 41, No. 5 (1995): 1439-1451.
[26] Chang, Yen-Ching. "Speeding Up Estimation of the Hurst Exponent by A Two-Stage Procedure from A Large to Small Range." Engineering Computations (2017).
[27] Chang, Yen-Ching. "An Efficient Maximum Likelihood Estimator for Two-Dimensional Fractional Brownian Motion." Fractionals (2020).
[28] Karatzas, Ioannis, And Steven E. Shreve. "Brownian Motion." In Brownian Motion and Stochastic Calculus, Pp. 47-127. Springer, New York, NY, 1998.
[29] Van Erven, Tim, And Peter Harremos. "Rényi Divergence and Kullback-Leibler Divergence." IEEE Transactions On Information Theory 60, No. 7 (2014): 3797-3820.
[30] Koutsoyiannis, Demetris. "The Hurst Phenomenon and Fractional Gaussian Noise Made Easy." Hydrological Sciences Journal 47, No. 4 (2002): 573-595.
[31] He, Xiaofei, Deng Cai, And Partha Niyogi. "Laplacian Score for Feature Selection." In Advances in Neural Information Processing Systems, Pp. 507-514. 2006.
[32] Ng, Andrew Y. "Feature Selection, L 1 Vs. L 2 Regularization, And Rotational Invariance." In Proceedings of the Twenty-First International Conference On Machine Learning, P. 78. 2004.
[33] Baldi, Pierre, And Peter J. Sadowski. "Understanding Dropout." In Advances in Neural Information Processing Systems, Pp. 2814-2822. 2013.
[34] Yao, Yuan, Lorenzo Rosasco, And Andrea Caponnetto. "On Early Stopping in Gradient Descent Learning." Constructive Approximation 26, No. 2 (2007): 289-315.
[35] Zhang, Tong, And Bin Yu. "Boosting with Early Stopping: Convergence and Consistency." The Annals of Statistics 33, No. 4 (2005): 1538-1579.
[36] Dietterich, Thomas G. "Ensemble Learning." The Handbook of Brain Theory and Neural Networks 2 (2002): 110-125.
[37] Sagi, Omer, And Lior Rokach. "Ensemble Learning: A Survey." Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery 8, No. 4 (2018): E1249.
[38] Momeni, Alexandre, Marc Thibault, And Olivier Gevaert. "Dropout-Enabled Ensemble Learning for Multi-Scale Biomedical Data." In International MICCAI Brainlesion Workshop, Pp. 407-415. Springer, Cham, 2018.
[39] Tsouros, Dimosthenis C., Stamatia Bibi, And Panagiotis G. Sarigiannidis. "A Review On UAV-Based Applications for Precision Agriculture." Information 10, No. 11 (2019): 349.
[40] Food and Agriculture Organization of the United Nations. The State of Food and Agriculture: Leveraging Food Systems for Inclusive Rural Transformation. 2017. Food and Agriculture Organization of the United Nations, 2017.
[41] De Janvry, Alain, Karen Macours, And Elisabeth Sadoulet. Learning for Adopting: Technology Adoption in Developing Country Agriculture. 2017.
[42] Lin, Tsung-Yi, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, And C. Lawrence Zitnick. "Microsoft Coco: Common Objects in Context." In European Conference On Computer Vision, Pp. 740-755. Springer, Cham, 2014.
[43] Redmon, Joseph, And Ali Farhadi. "Yolov3: An Incremental Improvement." Arxiv Preprint Arxiv:1804.02767 (2018).
[44] Held, David, Sebastian Thrun, And Silvio Savarese. "Learning to Track at 100 Fps with Deep Regression Networks." In European Conference On Computer Vision, Pp. 749-765. Springer, Cham, 2016.
[45] Wojke, Nicolai, Alex Bewley, And Dietrich Paulus. "Simple Online and Realtime Tracking with A Deep Association Metric." In 2017 IEEE International Conference On Image Processing (ICIP), Pp. 3645-3649. IEEE, 2017.
[46] Merwe, Rudolph Van Der, And Eric A. Wan. "Sigma-Point Kalman Filters for Integrated Navigation." In Proceedings of The 60th Annual Meeting of the Institute of Navigation (2004), Pp. 641-654. 2004.
[47] Zhao, Zhong-Qiu, Peng Zheng, Shou-Tao Xu, And Xindong Wu. "Object Detection with Deep Learning: A Review." IEEE Transactions On Neural Networks and Learning Systems 30, No. 11 (2019): 3212-3232.
[48] Lowe, David G. "Distinctive Image Features from Scale-Invariant Keypoints." International Journal of Computer Vision 60, No. 2 (2004): 91-110.
[49] Bay, Herbert, Tinne Tuytelaars, And Luc Van Gool. "Surf: Speeded Up Robust Features." In European Conference On Computer Vision, Pp. 404-417. Springer, Berlin, Heidelberg, 2006.
[50] Dalal, Navneet, And Bill Triggs. "Histograms of Oriented Gradients for Human Detection." In 2005 IEEE Computer Society Conference On Computer Vision and Pattern Recognition (CVPR'05), Vol. 1, Pp. 886-893. IEEE, 2005.
[51] Lienhart, Rainer, And Jochen Maydt. "An Extended Set of Haar-Like Features for Rapid Object Detection." In Proceedings. International Conference On Image Processing, Vol. 1, Pp. I-I. IEEE, 2002.
[52] Liu, Li, Wanli Ouyang, Xiaogang Wang, Paul Fieguth, Jie Chen, Xinwang Liu, And Matti Pietikäinen. "Deep Learning for Generic Object Detection: A Survey." International Journal of Computer Vision 128, No. 2 (2020): 261-318.
[53] Li, Ruibo, Ke Xian, Chunhua Shen, Zhiguo Cao, Hao Lu, And Lingxiao Hang. "Deep Attention-Based Classification Network for Robust Depth Prediction." In Asian Conference On Computer Vision, Pp. 663-678. Springer, Cham, 2018.
[54] Bochinski, Erik, Tobias Senst, And Thomas Sikora. "Extending IOU Based Multi-Object Tracking by Visual Information." In 2018 15th IEEE International Conference On Advanced Video and Signal Based Surveillance (AVSS), Pp. 1-6. IEEE, 2018.
[55] AI Platform for Autonomous Machines - NVIDIA Jetson AGX Xavier, 2018, [Online] Available: Https://Www.Nvidia.Com/En-Us/Autonomous-Machines/Jetson-Agx-Xavier/.
[56] Debauche, Olivier, Saïd Mahmoudi, Meryem Elmoulat, Sidi Ahmed Mahmoudi, Pierre Manneback, And Frédéric Lebeau. "Edge AI-Iot Pivot Irrigation, Plant Diseases, And Pests Identification." Procedia Computer Science 177 (2020): 40-48.
[57] Shi, Yuanming, Kai Yang, Tao Jiang, Jun Zhang, And Khaled B. Letaief. "Communication-Efficient Edge AI: Algorithms and Systems." Arxiv Preprint Arxiv:2002.09668 (2020).
[58] Kurakin, Alexey, Ian Goodfellow, And Samy Bengio. "Adversarial Examples In The Physical World." Arxiv Preprint Arxiv:1607.02533 (2016).
[59] Li, Bai, Changyou Chen, Wenlin Wang, And Lawrence Carin. "Second-Order Adversarial Attack and Certifiable Robustness." (2018).
[60] Tramèr, Florian, Alexey Kurakin, Nicolas Papernot, Ian Goodfellow, Dan Boneh, And Patrick Mcdaniel. "Ensemble Adversarial Training: Attacks and Defenses." Arxiv Preprint Arxiv:1705.07204 (2017).
[61] Abadi, Martín, Paul Barham, Jianmin Chen, Zhifeng Chen, Andy Davis, Jeffrey Dean, Matthieu Devin Et Al. "Tensorflow: A System for Large-Scale Machine Learning." In 12th {USENIX} Symposium On Operating Systems Design and Implementation ({OSDI} 16), Pp. 265-283. 2016.
[62] Szegedy, Christian, Wojciech Zaremba, Ilya Sutskever, Joan Bruna, Dumitru Erhan, Ian Goodfellow, And Rob Fergus. "Intriguing Properties of Neural Networks." Arxiv Preprint Arxiv:1312.6199 (2013).
[63] Carlini, Nicholas, And David Wagner. "Towards Evaluating the Robustness of Neural Networks." In 2017 Ieee Symposium On Security and Privacy (Sp), Pp. 39-57. IEEE, 2017.
[64] Xiao, Chaowei, Jun-Yan Zhu, Bo Li, Warren He, Mingyan Liu, And Dawn Song. "Spatially Transformed Adversarial Examples." Arxiv Preprint Arxiv:1801.02612 (2018).
[65] Wei, Mingwei, And David J. Schwab. "How Noise Affects the Hessian Spectrum in Overparameterized Neural Networks." Arxiv Preprint Arxiv:1910.00195 (2019).
[66] Uesato, Jonathan, Brendan O'Donoghue, Aaron Van Den Oord, And Pushmeet Kohli. "Adversarial Risk and The Dangers of Evaluating Against Weak Attacks." Arxiv Preprint Arxiv:1802.05666 (2018).
[67] Liu, Yujie, Shuai Mao, Xiang Mei, Tao Yang, And Xuran Zhao. "Sensitivity of Adversarial Perturbation in Fast Gradient Sign Method." In 2019 IEEE Symposium Series On Computational Intelligence (SSCI), Pp. 433-436. IEEE, 2019.
[68] Feinman, Reuben, Ryan R. Curtin, Saurabh Shintre, And Andrew B. Gardner. "Detecting Adversarial Samples from Artifacts." Arxiv Preprint Arxiv:1703.00410 (2017).
[69] Athalye, Anish, Nicholas Carlini, And David Wagner. "Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples." Arxiv Preprint Arxiv:1802.00420 (2018).
[70] Weng, Lily, Pin-Yu Chen, Lam Nguyen, Mark Squillante, Akhilan Boopathy, Ivan Oseledets, And Luca Daniel. "PROVEN: Verifying Robustness of Neural Networks with A Probabilistic Approach." In International Conference On Machine Learning, Pp. 6727-6736. PMLR, 2019.
[71] Zhang, Hongyang, Yaodong Yu, Jiantao Jiao, Eric P. Xing, Laurent El Ghaoui, And Michael I. Jordan. "Theoretically Principled Trade-Off Between Robustness and Accuracy." Arxiv Preprint Arxiv:1901.08573 (2019).
[72] Hao-Chen, Han Xu Yao Ma, Liu Debayan Deb, Hui Liu Ji-Liang Tang Anil, And K. Jain. "Adversarial Attacks and Defenses in Images, Graphs and Text: A Review." International Journal of Automation and Computing 17, No. 2 (2020): 151-178.
[73] Tzutalin. LabelImg. Git code (2015). https://github.com/tzutalin/labelImg