臺灣博碩士論文加值系統

近年來，溺水事件引起了社會的高度關注，因其對公共安全造成的潛在威脅。隨著科技的不斷發展，我們意識到傳統的監測方法已經無法完全滿足現代社會對於溺水事件處理的需求。因此，我們利用深度學習和電腦視覺技術來識別和監控溺水事件，以提高救援效率和溺水者的生存率。溺水者通常在遇到危險時無法發出清晰的求救呼叫或展現明顯的掙扎，這對於救援人員增加了挑戰。
本論文的目標為利用電腦視覺技術，分析溺水行為的特徵，進而快速準確地辨識潛在的溺水事件。本論文所面臨的主要挑戰在於溺水數據集的數量不足，這將對模型訓練的準確性和可靠性產生影響。因此，我們與游泳池現場救生人員合作，模擬各種可能溺水情境，以獲得更貼近實際的溺水行為數據。這使得我們的模型更加貼近實際應用場景，提高了救援的效率和溺水者的生存機會。

In recent years, drowning incidents have attracted widespread societal attention due to the potential threats they pose to public safety. With the continuous advancement of technology, we have realized that traditional monitoring methods are no longer sufficient to meet modern society's demands for handling drowning incidents. Consequently, we employ deep learning and computer vision techniques to identify and monitor drowning events, aiming to improve rescue efficiency and the survival rate of drowning victims. Drowning individuals are typically unable to issue clear distress calls or exhibit obvious struggles when in danger, posing challenges for rescue personnel. This paper aims to leverage computer vision techniques to analyze the characteristics of drowning behavior, thereby rapidly and accurately identifying potential drowning incidents.
The primary challenge faced in this paper is the insufficient quantity of drowning datasets, which can impact the accuracy and reliability of model training. Therefore, we collaborate with on-site lifeguards at swimming pools to simulate various potential drowning scenarios, enabling us to obtain more realistic drowning behavior data. This ensures that our model is better aligned with real-world application scenarios, improving rescue efficiency and increasing the chances of survival for drowning victims.

摘要...i
Abstract...ii
誌謝...iii
目錄...iv
表目錄...vi
圖目錄...vii
1 第一章 緒論...1
1.1 研究動機...1
1.2 研究目的...1
1.3 論文架構...2
2 第二章 機器學習...3
2.1 監督式學習...3
2.2 非監督式學習...3
2.3 自監督式學習...4
2.4 強化式學習...4
3 第三章 神經網路架構及原理...6
3.1 類神經網路...6
3.1.1 神經網路架構...6
3.1.2 神經網路原理...6
3.1.3 神經元(Neuron)...7
3.1.4 權重(Weight)...8
3.1.5 激勵函數(Activation Function)...8
3.1.6 優化器(Optimizer)...11
3.1.7 前向傳播(Forword Propagation)...12
3.1.8 反向傳播(Backpropagation)...12
3.1.9 損失函數(Loss Function)...12
3.1.10 訓練(Training)...13
3.2 卷積神經網路...14
3.2.1 Convolutional卷積...14
3.2.2 Padding填充...14
3.2.3 Pooling Layer池化層...15
3.3 Visual Geometry Group 16-layer (VGG16)...17
3.4 殘差卷積神經網路(ResNet)...18
4 第四章 研究內容與實驗架構...19
4.1 分類器...19
4.2 姿態估計...23
4.2.1 Backbone網路...25
4.2.2 空間金字塔池化（Spatial Pyramid Pooling, SPP）...25
4.2.3 YOLO Head...26
4.2.4 YOLOv7 Loss...27
4.3 Jetson Nano...28
4.4 實驗架構...29
5 第五章 實驗結果與討論...30
5.1 實驗結果...30
5.1.1 判定結果-分類器為溺水，姿態估計為溺水...31
5.1.2 判定結果-分類器為正常...33
5.1.3 判定結果-分類器為溺水，姿態估計為正常...36
5.1.4 判定結果-分類器為溺水，姿態估計無法確定...36
5.1.5 YOLOv7使用影片的姿態估計結果...37
6 第六章 結論...43
6.1 論文結論...43
6.2 未來研究...43
參考文獻 (References)...44
Extended Abstract...46

[1] M. Pushpavalli, P. Abirami, P. Sivagami, M. Kavitha, C. Bhuvaneswari, and W. A. Memala, "Creation of a Drowning Prevention System for Kids Learning in Pool using Computational Technology," in 2023 International Conference on Artificial Intelligence and Knowledge Discovery in Concurrent Engineering (ICECONF), 5-7 Jan. 2023 2023, pp. 1-5, doi: 10.1109/ICECONF57129.2023.10083598.
[2] M. Rakotondraibe, T. Fang, and J. Saniie, "Drowning Recognition for Ocean Surveillance using Computer Vision and Drone Control," in 2023 IEEE International Conference on Electro Information Technology (eIT), 18-20 May 2023 2023, pp. 394-398, doi: 10.1109/eIT57321.2023.10187281.
[3] B. Ganesamoorthy, R. M, A. K, and M. T, "Automated Drowning Detection and Security in Swimming Pool With IoT Server," in 2023 International Conference on System, Computation, Automation and Networking (ICSCAN), 17-18 Nov. 2023 2023, pp. 1-5, doi: 10.1109/ICSCAN58655.2023.10395724.
[4] C. W. Li and C. M. Huang, "Application of Deep Learning for the Classification of Drowning Incidents," in 2023 IEEE 5th Eurasia Conference on IOT, Communication and Engineering (ECICE), 27-29 Oct. 2023 2023, pp. 340-342, doi: 10.1109/ECICE59523.2023.10383060.
[5] D. A. R. Ellen, P. Kristalina, M. Z. S. Hadi, and A. Patriarso, "Effective Searching of Drowning Victims in the River using Deep Learning Method and Underwater Drone," in 2023 International Electronics Symposium (IES), 8-10 Aug. 2023 2023, pp. 569-574, doi: 10.1109/IES59143.2023.10242589.
[6] R. Hongyo, Y. Egashira, and K. Yamaguchi, "Deep Neural Network Based Predistorter with ReLU Activation for Doherty Power Amplifiers," in 2018 Asia-Pacific Microwave Conference (APMC), 6-9 Nov. 2018 2018, pp. 959-961, doi: 10.23919/APMC.2018.8617612.
[7] L. K. Kalyanam and S. Katkoori, "Unstructured Pruning for Multi-Layer Perceptrons with Tanh Activation," in 2023 IEEE International Symposium on Smart Electronic Systems (iSES), 18-20 Dec. 2023 2023, pp. 69-74, doi: 10.1109/iSES58672.2023.00025.
[8] S. Xing and C. Wu, "Implementation of A Neuron Using Sigmoid Activation Function with CMOS," in 2020 IEEE 5th International Conference on Integrated Circuits and Microsystems (ICICM), 23-25 Oct. 2020 2020, pp. 201-204, doi: 10.1109/ICICM50929.2020.9292239.
[9] J. Xu, Z. Li, B. Du, M. Zhang, and J. Liu, "Reluplex made more practical: Leaky ReLU," in 2020 IEEE Symposium on Computers and Communications (ISCC), 7-10 July 2020 2020, pp. 1-7, doi: 10.1109/ISCC50000.2020.9219587.
[10]Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner, "Gradient-based learning applied to document recognition," Proceedings of the IEEE, vol. 86, no. 11, pp. 2278-2324, 1998, doi: 10.1109/5.726791.
[11] B. A. Dabwan, F. A. Olayah, H. T. Halawani, A. M. Mashraqi, Y. A. Abdelrahman, and A. F. Shamsan, "Classification of Sign language Using VGG16 and VGG19," in 2023 7th International Symposium on Innovative Approaches in Smart Technologies (ISAS), 23-25 Nov. 2023 2023, pp. 1-6, doi: 10.1109/ISAS60782.2023.10391673.
[12] K. He, X. Zhang, S. Ren, and J. Sun, "Deep Residual Learning for Image Recognition," in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 27-30 June 2016 2016, pp. 770-778, doi: 10.1109/CVPR.2016.90.
[13]K. Simonyan and A. Zisserman, "Very Deep Convolutional Networks for Large-Scale Image Recognition," CoRR, vol. abs/1409.1556, 2014.
[14] H. Zhang, A. Xiong, L. Lai, C. Chen, and J. Liang, "AMME-YOLOv7: Improved YOLOv7 Based on Attention Mechanism and Multiscale Expansion for Electric Vehicle Driver and Passenger Helmet Wearing Detection," in 2023 IEEE International Conference on Smart Internet of Things (SmartIoT), 25-27 Aug. 2023 2023, pp. 223-227, doi: 10.1109/SmartIoT58732.2023.00039.
[15] B. Bai, L. Chen, and X. Li, "Improved YOLOv7 Fusion Detection Line for Swimming Pool Drowning Detection," in 2023 IEEE 16th International Conference on Electronic Measurement & Instruments (ICEMI), 9-11 Aug. 2023 2023, pp. 129-134, doi: 10.1109/ICEMI59194.2023.10270676.
[16]B. Li, B. Liu, S. Li, and H. Liu, "Underwater target detection based on improved YOLOv7," ArXiv, vol. abs/2302.06939, 2023.
[17] A. Jose, R. D. Lopez, I. Heisterklaus, and M. Wien, "Pyramid Pooling of Convolutional Feature Maps for Image Retrieval," in 2018 25th IEEE International Conference on Image Processing (ICIP), 7-10 Oct. 2018 2018, pp. 480-484, doi: 10.1109/ICIP.2018.8451361.
[18] X. Zhang, Y. Zhang, M. Hu, and X. Ju, "Insulator defect detection based on YOLO and SPP-Net," in 2020 International Conference on Big Data & Artificial Intelligence & Software Engineering (ICBASE), 30 Oct.-1 Nov. 2020 2020, pp. 403-407, doi: 10.1109/ICBASE51474.2020.00092.