人工智慧這名詞現在已四處可見，完全不陌生。各個領域未來都將與人工智慧做結合，許多領 域逐漸導入機器學習，類神經網路更是成為主要的研究方向。
現代城市面臨垃圾管理的挑戰，垃圾丟棄問題對環境和社會造成了嚴重的影響，因此一個有 效的偵測系統可以協助更有效的監控和解決垃圾處理的相關問題。
本研究的題目將著重如何利用基於 YOLO (You Only Look Once) 模型以有效辨識並定位環境 中亂丟垃圾的行為。使用深度學習技術，特別是卷積神經網路 (CNN) 和 YOLO 演算法，結合對 象檢測與圖像分類的強大能力，對影像進行即時目標檢測，並透過訓練模型識別特定的垃圾物品， 不僅能迅速辨識垃圾物品，還能提供物品所在的確定位置，若為人為丟起的垃圾物品，則可以記 錄當下丟棄行為人亦或者由車輛上掉落的物品即可紀錄丟棄物品之車輛車牌，本研究可應用於城 市、公園或其他公共場所，協助監控和防範亂丟垃圾現象，進一步提升城市管理效率及環境管理 的自動化。

The term “artificial intelligence” is now ubiquitous and familiar, with its integration expected
across various fields in the future. Many domains are gradually adopting machine learning, with neural networks becoming a primary focus of research.
Modern cities face challenges in waste management, as improper disposal significantly impacts both the environment and society. Therefore, an effective detection system can assist in monitoring and addressing waste management issues more efficiently.
This study focuses on utilizing the YOLO (You Only Look Once) model to effectively identify and locate littering behavior in the environment. By employing deep learning techniques, particularly convolutional neural networks (CNNs) and the YOLO algorithm, the powerful capabilities of object detection and image classification are combined. This allows for real-time target detection in images, and through training the model to recognize specific litter items, is can swiftly identify them and provide their precise locations. If the litter is the result of human behavior, the system can record the act of littering; if it’s debris falling from vehicles, it can capture the vehicle’s license plate. This research can be applied in cities, parks, or other public areas to assist in monitoring and preventing littering,
thus ,enhancing urban management efficiency and environmental automation.

目錄
致謝--------------------------------------------------------------------I 
摘要--------------------------------------------------------------------II 
ABSTRACT---------------------------------------------------------------III 
目錄--------------------------------------------------------------------IV 
圖目錄------------------------------------------------------------------V 
表目錄------------------------------------------------------------------VII 
第一章 緒論--------------------------------------------------------------1
1.1 研究背景-------------------------------------------------------------1 
1.2 研究動機-------------------------------------------------------------2 
1.3 研究目的-------------------------------------------------------------3 
1.4 論文架構-------------------------------------------------------------4
第二章 相關文獻-----------------------------------------------------------5 
2.1 YOLO 物件辨識模型-----------------------------------------------------5 
2.1.1 YOLOv1------------------------------------------------------------5 
2.1.2 YOLOv2------------------------------------------------------------7 
2.1.3 YOLOv3------------------------------------------------------------8 
2.1.4 YOLOv4------------------------------------------------------------10 
2.1.5 YOLOv8------------------------------------------------------------11 
第三章 研究方法------------------------------------------------------------14 
3.1 資料搜集--------------------------------------------------------------15 
3.2 素材標註--------------------------------------------------------------15 
3.3 訓練-----------------------------------------------------------------16 
3.3.1 資料擴增------------------------------------------------------------16 
3.3.2 模型訓練------------------------------------------------------------17 
3.4 偵測與識別------------------------------------------------------------22 
3.4.1 物件偵測------------------------------------------------------------22 
3.4.2 多物件排序-----------------------------------------------------------23
3.4.3 行為判定------------------------------------------------------------23 
第四章 實驗結果與分析-------------------------------------------------------25 
4.1 量化評估方式-----------------------------------------------------------25 
4.1.1 混淆矩陣------------------------------------------------------------25 
4.1.2 mAP 及 TPR 與 FRR---------------------------------------------------25 
4.2 亂丟垃圾行為辨識實驗結果-------------------------------------------------28
4.2.1 實驗結果--------------------------------------------------------------29  
第五章 結論與未來展望--------------------------------------------------------34 
參考文獻-------------------------------------------------------------------35


圖目錄
圖 2.1 YOLO 版本時間軸 -------------------------------------------------------5 
圖 2.2 YOLO 辨識過程 ---------------------------------------------------------6 
圖 2.3 YOLOv1 架構 ----------------------------------------------------------6 
圖 2.4 錨框 -----------------------------------------------------------------7 
圖 2.5 YOLOv4 主要架構-------------------------------------------------------10 
圖 2.6 CSPDarknet-53 架構----------------------------------------------------11 
圖 2.7 YOLOv8 架構 ----------------------------------------------------------13 
圖 3.1 研究流程架構圖 --------------------------------------------------------14 
圖 3.2 使用 LabelImg 進行標注-------------------------------------------------16 
圖 3.3 分數與置信度關係圖------------------------------------------------------19 
圖 3.4 精確率與召回率關係圖----------------------------------------------------20 
圖 3.5 召回率與置信度關係圖----------------------------------------------------20 
圖 3.6 YOLOv8 歸一化混淆矩陣分析圖---------------------------------------------21
圖 3.7 精確度與智信關係圖------------------------------------------------------21 
圖 4.1 AP 計算流程-----------------------------------------------------------26 
圖 4.2 Precision 與 Recall--------------------------------------------------27 
圖 4.3 夜市辨識畫面-----------------------------------------------------------29 
圖 4.4 低畫質監視器辨識畫面----------------------------------------------------30
圖 4.5 網路影片辨識畫面--------------------------------------------------------31 
圖 4.6 新聞辨識畫面-----------------------------------------------------------33

表目錄
表 2.1 Darknet-19 ----------------------------------------------------------8 
表 2.2 Darknet-53 ----------------------------------------------------------9 
表 3.1 資料擴增參數-----------------------------------------------------------17 
表 3.2 訓練設備配置-----------------------------------------------------------17 
表 3.3 參數設定--------------------------------------------------------------18 
表 3.4 YOLOv8 驗證結果統計----------------------------------------------------18 
表 3.5 YOLOv8 的各項驗證結果--------------------------------------------------19
表 3.6 Yolo 識別回傳資訊------------------------------------------------------22 
表 4.1 混淆矩陣 -------------------------------------------------------------25

參考文獻
[1] J. Redmon, S. Divvala, R. Girshick and A. Farhadi. You only look once: Unified, Real-time object detection. arXiv:1506.02640, 2015
[2] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke and A. Rabinovich. Going deeper with convolutions. arXiv:1409.4842, 2014
[3] Terven, J.; Córdova-Esparza, D.-M.; Romero-González, J.-A. A Comprehensive Review of YOLO Architectures in Computer Vision: From YOLOv1 to YOLOv8 and YOLO-NAS. Mach. Learn. Knowl. Extr. 2023, 5, 1680-1716.
[4] K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recognition. arXiv:1409.1556, 2015
[5] K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recognition. arXiv:1409.1556, 2015
[6] T. Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan and S. Belongie(2016), “Feature Pyramid Networks for Object Detection,” Computer Vision and Pattern Recogntion, arXiv.1612
[7] A. Bochkovskiy, C. Wang and H-M. Liao. YOLOv4: Optimal speed and accuracy of object detection. arXiv:2004.10934, 2020
[8] C-Y. Wang, H-M. Liao, I-H. Yeh, Y-H. Wu, P-Y. Chen and J-W. Hsieh. CSPNet: A new backbone that can enhance learning capability of CNN. arXiv:1911.11929, 2019
[9] Dai,J.;Qi,H.;Xiong,Y.;Li,Y.;Zhang,G.;Hu,H.;Wei,Y.Deformableconvolutional networks. In Proceedings of the IEEE International Conference on Computer Vision, Venice, Italy, 22–29 October 2017; pp. 764–773.
[10]Zheng, Z.; Wang, P.; Liu, W.; Li, J.; Ye, R.; Ren, D. Distance-IoU loss: Faster and better learning for bounding box regression. In Proceedings of the AAAI Conference on Artificial Intelligence, New York, NY, USA, 7–12 February 2020; Volume 34, pp. 12993–13000.
[11]Li, X.; Wang, W.; Wu, L.; Chen, S.; Hu, X.; Li, J.; Tang, J.; Yang, J. Generalized focal loss: Learning qualified and distributed bounding boxes for dense object detection. Adv. Neural Inf. Process. Syst. 2020, 33, 21002–21012.