臺灣博碩士論文加值系統

本研究開發的傷口檢測系統屬於醫療輔助診斷系統的一種，傷口的成因不同，且傷口復原的過程複雜，因此過去傷口的診斷與追蹤，多依靠醫護人員主觀及專業的判斷，或透過人工紀錄數據及拍照攝影，當醫護人員的經驗不足，或記錄時產生的誤差及疏失，就可能會造成錯誤的診斷而延誤治療，因此透過影像分析技術，協助醫護人員以更客觀且更準確的的條件攫取傷口資訊，是有其必要性。
我們以深度學習演算法中的物件偵測模型YOLOv8及Mask R-CNN，做為系統建置的核心，從傷口影像資料蒐集到影像標註，然後使用遷移式學習法來減少模型訓練資源並達更好的效果，透過資料量擴增實驗、13種數據增強實驗及參數最佳化實驗，來提升模型性能。
最終本研究所開發的傷口檢測系統的最佳化模型，YOLOv8的實例分割指標mAP50可達0.992，推理時間為9.9ms，而Mask R-CNN的實例分割指標mAP50為0.87，推理時間為44ms，Mask R-CNN在少量的資料集的情況下，雖然也能有不錯的檢測效果，但近年來快速發展的YOLO家族系列，已經成為性能及速度兼具的模型，對於傷口檢測系統建置而言是更合適的選擇。

The wound detection system developed in this study belongs to a type of medical auxiliary diagnostic system. Due to the different causes of wounds and the complex process of wound healing, in the past, the diagnosis and tracking of wounds mostly relied on the subjective and professional judgment of medical personnel, or through manual recording of data or taking photographs. When medical personnel lack experience or errors and omissions occur during recording, it may lead to incorrect diagnosis and delay in treatment. Therefore, it is necessary to assist medical personnel in capturing wound information under more objective and accurate conditions through image analysis technology.
We used the object detection models YOLOv8 and Mask R-CNN in deep learning algorithms as the core of the system construction, from collecting wound image data to image annotation, and then using transfer learning to reduce model training resources and achieve better results. Through experiments on data augmentation, 13 types of data augmentation, and parameter optimization, the performance of the model is improved.
Ultimately, the optimized model of the wound detection system developed in this study has an instance segmentation index mAP50 of 0.992 and an inference time of 9.9ms for YOLOv8, while Mask R-CNN has an instance segmentation index mAP50 of 0.87 and an inference time of 44ms. Although Mask R-CNN also shows good detection results with a small amount of data, in recent years, the rapidly developing YOLO family series has become a model with both performance and speed. It is a more suitable choice for the construction of wound detection systems.

摘要 i
ABSTRACT ii
誌謝 iv
目錄 v
表目錄 ix
圖目錄 xi
第一章 緒論 1
1.1 研究背景與動機 1
1.2 研究目的 2
1.3 文獻回顧 3
1.4 論文貢獻 4
1.5 論文架構 5
第二章 相關技術探討 6
2.1 物件偵測演算法 6
2.2 One-stage Detector vs Two-stage Detector 7
2.3 卷積神經網路 8
2.3.1 卷積層 8
2.3.2 池化層 9
2.3.3 全連接層 10
2.4 R-CNN 11
2.4.1 選擇性搜索 12
2.4.2 非極大值抑制 13
2.4.3 支持向量機 13
2.4.4 邊框回歸 14
2.5 Fast R-CNN 15
2.6 Faster R-CNN 15
2.6.1 Anchors 16
2.6.2 RPN 17
2.7 Mask R-CNN 18
2.7.1 FPN 18
2.7.2 RoI Align 19
2.7.3 Mask Branch 20
2.8 YOLO 20
2.8.1 YOLOv1 20
2.8.2 YOLOv2 22
2.8.3 YOLOv3 24
2.8.4 YOLOv4 24
2.8.5 YOLOv5 26
2.8.6 YOLOv6 27
2.8.7 YOLOv7 27
2.8.8 YOLOv8 30
第三章 研究方法 32
3.1 傷口影像資料集 33
3.1.1 傷口影像資料蒐集 33
3.1.2 傷口影像資料分類 33
3.1.3 傷口影像資料前處理 36
3.2 影像標記 36
3.3 數據增強 37
3.4 建立模型 41
3.4.1 Pytorch 41
3.4.2 Colab 41
3.4.3 Detectron2 41
3.4.4 Ultralytics 42
3.5 參數設置及最佳化 42
3.6 訓練模型 42
3.6.1 監督式學習 42
3.6.2 遷移式學習 43
3.7 損失函數 43
3.7.1 均方誤差 44
3.7.2 交叉熵 44
3.8 模型評價指標 44
3.8.1 IoU 45
3.8.2 Confusion Matrix 45
3.8.3 Precision 46
3.8.4 Recall 46
3.8.5 Accuracy 46
3.8.6 F1-Score 46
3.8.7 AP 47
3.8.8 mAP 47
3.8.9 mAP50與mAP50-95 47
第四章 研究結果與分析 48
4.1 實驗環境 48
4.2 預訓練模型 50
4.3 訓練參數設置 51
4.4 模型資料量擴增之實驗結果與探討 51
4.4.1 Mask R-CNN資料量擴增之實驗結果與分析 51
4.4.2 YOLOv8資料量擴增之實驗結果與分析 54
4.4.3 YOLOv8與Mask R-CNN資料量擴增實驗比較 57
4.5 模型數據增強之實驗結果與探討 58
4.5.1 Mask R-CNN數據增強之實驗結果與分析 58
4.5.2 YOLOv8數據增強之實驗結果與分析 60
4.5.3 YOLOv8與Mask R-CNN數據增強實驗比較 62
4.6 模型參數最佳化之實驗結果與探討 63
4.6.1 Mask R-CNN參數最佳化之實驗結果與分析 63
4.6.2 YOLOv8參數最佳化之實驗結果與分析 66
4.6.3 YOLOv8與Mask R-CNN參數最佳化實驗比較 68
4.7 傷口檢測系統最佳化模型展示 69
4.7.1 Mask R-CNN最佳化模型 69
4.7.2 YOLOv8最佳化模型 72
第五章 結論與未來展望 76
5.1 結論 76
5.2 未來展望 78
參考文獻 79

[1]D. S. Kermany et al., "Identifying medical diagnoses and treatable diseases by image-based deep learning," cell, vol. 172, no. 5, pp. 1122-1131. e9, 2018.
[2]M. Rodrigues, N. Kosaric, C. A. Bonham, and G. C. Gurtner, "Wound healing: a cellular perspective," Physiological reviews, vol. 99, no. 1, pp. 665-706, 2019.
[3]H. Sorg, D. J. Tilkorn, S. Hager, J. Hauser, and U. Mirastschijski, "Skin wound healing: an update on the current knowledge and concepts," European surgical research, vol. 58, no. 1-2, pp. 81-94, 2017.
[4]C. T. Hess, "The art of skin and wound care documentation," Advances in skin & wound care, vol. 18, no. 1, pp. 43-53, 2005.
[5]H. Khalil, M. Cullen, H. Chambers, N. Steers, and J. Walker, "Implementation of a successful electronic wound documentation system in rural Victoria, Australia: a subject of collaboration and community engagement," International wound journal, vol. 11, no. 3, pp. 314-318, 2014.
[6]C. K. Sen, "Human wounds and its burden: an updated compendium of estimates," vol. 8, ed: Mary Ann Liebert, Inc., publishers 140 Huguenot Street, 3rd Floor New …, 2019, pp. 39-48.
[7]K. He, G. Gkioxari, P. Dollár, and R. Girshick, "Mask r-cnn," in Proceedings of the IEEE international conference on computer vision, 2017, pp. 2961-2969.
[8]M. F. A. Fauzi, I. Khansa, K. Catignani, G. Gordillo, C. K. Sen, and M. N. Gurcan, "Computerized segmentation and measurement of chronic wound images," Computers in biology and medicine, vol. 60, pp. 74-85, 2015.
[9]M. Goyal, N. D. Reeves, S. Rajbhandari, and M. H. Yap, "Robust methods for real-time diabetic foot ulcer detection and localization on mobile devices," IEEE journal of biomedical and health informatics, vol. 23, no. 4, pp. 1730-1741, 2018.
[10]A. Abubakar, H. Ugail, and A. M. Bukar, "Can machine learning be used to discriminate between burns and pressure ulcer?," in Intelligent Systems and Applications: Proceedings of the 2019 Intelligent Systems Conference (IntelliSys) Volume 2, 2020: Springer, pp. 870-880.
[11]M. Cullell-Dalmau, M. Otero-Viñas, and C. Manzo, "Research techniques made simple: deep learning for the classification of dermatological images," Journal of Investigative Dermatology, vol. 140, no. 3, pp. 507-514. e1, 2020.
[12]C. Wang et al., "Fully automatic wound segmentation with deep convolutional neural networks," Scientific reports, vol. 10, no. 1, p. 21897, 2020.
[13]G. Scebba et al., "Detect-and-segment: A deep learning approach to automate wound image segmentation," Informatics in Medicine Unlocked, vol. 29, p. 100884, 2022.
[14]D. Y. Chino, L. C. Scabora, M. T. Cazzolato, A. E. Jorge, C. Traina-Jr, and A. J. Traina, "Segmenting skin ulcers and measuring the wound area using deep convolutional networks," Computer methods and programs in biomedicine, vol. 191, p. 105376, 2020.
[15]H. Carrión, M. Jafari, M. D. Bagood, H.-y. Yang, R. R. Isseroff, and M. Gomez, "Automatic wound detection and size estimation using deep learning algorithms," PLoS computational biology, vol. 18, no. 3, p. e1009852, 2022.
[16]A. Arnab and P. H. Torr, "Pixelwise instance segmentation with a dynamically instantiated network," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 441-450.
[17]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.
[18]R. Girshick, J. Donahue, T. Darrell, and J. Malik, "Rich feature hierarchies for accurate object detection and semantic segmentation," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2014, pp. 580-587.
[19]B. Liu, Z.-F. Hao, and X.-W. Yang, "Nesting support vector machinte for muti-classification [machinte read machine]," in 2005 International Conference on Machine Learning and Cybernetics, 2005, vol. 7: IEEE, pp. 4220-4225.
[20]D. Wang, X. Chen, H. Yi, and F. Zhao, "Improvement of non-maximum suppression in RGB-D object detection," IEEE Access, vol. 7, pp. 144134-144143, 2019.
[21]R. Girshick, "Fast r-cnn," in Proceedings of the IEEE international conference on computer vision, 2015, pp. 1440-1448.
[22]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, vol. 28, 2015.
[23]X. Xu et al., "Crack detection and comparison study based on faster R-CNN and mask R-CNN," Sensors, vol. 22, no. 3, p. 1215, 2022.
[24]T.-Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan, and S. Belongie, "Feature pyramid networks for object detection," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 2117-2125.
[25]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, 2016, pp. 779-788.
[26]J. Redmon and A. Farhadi, "YOLO9000: better, faster, stronger," in Proceedings of the IEEE conference on computer vision and pattern recognition, 2017, pp. 7263-7271.
[27]J. Redmon and A. Farhadi, "Yolov3: An incremental improvement," arXiv preprint arXiv:1804.02767, 2018.
[28]A. Bochkovskiy, C.-Y. Wang, and H.-Y. M. Liao, "Yolov4: Optimal speed and accuracy of object detection," arXiv preprint arXiv:2004.10934, 2020.
[29]J. Jiang, X. Fu, R. Qin, X. Wang, and Z. Ma, "High-speed lightweight ship detection algorithm based on YOLO-v4 for three-channels RGB SAR image," Remote Sensing, vol. 13, no. 10, p. 1909, 2021.
[30]H. Liu, F. Sun, J. Gu, and L. Deng, "Sf-yolov5: A lightweight small object detection algorithm based on improved feature fusion mode," Sensors, vol. 22, no. 15, p. 5817, 2022.
[31]C. Li et al., "YOLOv6: A single-stage object detection framework for industrial applications," arXiv preprint arXiv:2209.02976, 2022.
[32]C.-Y. Wang, A. Bochkovskiy, and H.-Y. M. Liao, "YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors," in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 7464-7475.
[33]K. Jiang et al., "An attention mechanism-improved YOLOv7 object detection algorithm for hemp duck count estimation," Agriculture, vol. 12, no. 10, p. 1659, 2022.
[34]H. Lou et al., "DC-YOLOv8: small-size object detection algorithm based on camera sensor," Electronics, vol. 12, no. 10, p. 2323, 2023.
[35]R. Padilla, S. L. Netto, and E. A. Da Silva, "A survey on performance metrics for object-detection algorithms," in 2020 international conference on systems, signals and image processing (IWSSIP), 2020: IEEE, pp. 237-242.