融合雙向強化學習與新型資料擴增技術於管線瑕疵檢測之創新研究

以作者查詢圖書館館藏

、以作者查詢臺灣博碩士

、以作者查詢全國書目

、勘誤回報

、線上人數：12

、訪客IP：18.119.121.38

姓名

陳以恩(Yi-En Chen) 查詢紙本館藏

畢業系所

機械工程學系

論文名稱

融合雙向強化學習與新型資料擴增技術於管線瑕疵檢測之創新研究

相關論文

★ 基於注意力和記憶動態融合的單物件追蹤方法	★ TCN-MAML: A TCN-based Model with Model-agnostic Meta-learning for Human-to-Human Interaction Recognition
★ Real-time Human Activity Recognition using WiFi Channel State Information	★ Accelerated Point Cloud Rendering and Density Enhancement via Depth Completion: An Improved ?3???? SLAM System Implementation

檔案

[Endnote RIS 格式]

[Bibtex 格式]

[相關文章]

[文章引用]

[完整記錄]

[館藏目錄]

至系統瀏覽論文 (2027-1-11以後開放)

摘要(中)

本研究針對工業管線鏽蝕檢測中的多樣化挑戰，提出了一套創新的解決方案，結合資料擴增技術與雙向強化學習機制，旨在提升檢測模型的精度、穩定性和泛化能力，以應對工業環境中複雜多變的鏽蝕特徵。
在資料擴增方面，本研究基於擴散模型（Diffusion Model）與其延伸技術（ControlNet）技術，生成高解析度且涵蓋多種形態的鏽蝕影像，並結合自建篩選模組，根據指定的鏽蝕等級進行特徵篩選與優化。此方法顯著擴充了訓練數據集的多樣性，並有效解決了原始數據集中樣本不足的問題。在多樣化場景下，生成的影像與實地拍攝影像相結合，使模型在小型、不規則及背景複雜的鏽蝕區域中表現出色。實驗結果顯示，結合此資料擴增技術後，YOLOv8 和 YOLOv11 模型的檢測準確度顯著提升，漏檢測率大幅降低，特別是在形狀不規則與背景干擾嚴重的鏽蝕區域中，檢測性能實現突破性改善。
此外，本研究引入了雙向強化學習機制，透過獎懲回饋機制實現模型策略的動態優化。在訓練初期，人為干預對模型檢測結果進行校正，幫助模型快速聚焦於鏽蝕區域。隨著訓練的進行，雙向強化學習逐漸取代人為干預，通過自適應調整檢測策略，增強了模型在多樣化工業場景中的適應能力，同時顯著提升了檢測結果的穩定性與準確性。綜合而言，本研究成功結合資料擴增技術與雙向強化學習機制，顯著提升了工業鏽蝕檢測模型的整體性能，並展現了兩者相結合的創新潛力與應用價值。

摘要(英)

This study addresses the challenges in industrial pipeline rust detection by proposing an innovative solution that combines data augmentation techniques with a bidirectional reinforcement learning mechanism. The goal is to enhance the precision, stability, and generalization capability of detection models to handle diverse and complex rust charac-teristics in industrial environments.
For data augmentation, this study utilizes Diffusion Models and ControlNet tech-nology to generate high-resolution rust images. A customized filtering module is integrat-ed to select and optimize features based on specified rust levels. This approach signifi-cantly expands the diversity of the training dataset and effectively resolves the issue of insufficient samples in the original dataset. By combining generated images with re-al-world captured images, the model demonstrates exceptional performance in detecting small, irregular, and complex rusted regions. Experimental results show that with this data augmentation technique, the detection accuracy of YOLOv8 and YOLOv11 models im-proved significantly, with a notable reduction in missed detections, particularly in irregular shapes and highly noisy backgrounds.
In addition, this study introduces a bidirectional reinforcement learning mechanism that dynamically optimizes model strategies through reward and penalty feedback. During the initial training phase, human intervention plays a critical role in correcting model de-tection results, helping the model focus on rust areas. As training progresses, the bidirec-tional reinforcement learning system gradually replaces human intervention by adaptively adjusting detection strategies. This enhancement strengthens the model’s adaptability to diverse industrial scenarios while significantly improving detection stability and accuracy. In summary, this study successfully integrates data augmentation techniques with a bidi-rectional reinforcement learning mechanism, substantially enhancing the overall perfor-mance of industrial rust detection models. It demonstrates the innovative potential and practical value of this combined approach in addressing complex rust detection tasks.

關鍵字(中)

★ 強化學習
★ 瑕疵檢測

關鍵字(英)

★ Reinforcement Learning
★ Defect Detection

論文目次

目錄
摘要 i
Abstract ii
致謝 iii
目錄 iv
圖目錄 vi
表目錄 viii
1. 緒論 1
1.1 研究背景 1
1.2 欲解決問題 2
1.3 論文貢獻 4
1.4 論文架構 5
2. 文獻回顧 6
2.1 物件偵測 7
2.1.1 You Only Look Once 7
2.1.2 Faster R-CNN 8
2.1.3 DETR 10
2.2 圖像分割 12
2.2.1 U-Net 12
2.2.2 Mask R-CNN 13
2.2.3 Segment Anything Model 15
2.3 圖像生成 16
2.3.1 Generative Adversarial Networks 16
2.3.2 Diffusion Model 17
2.3.3 ControlNet 18
2.4 強化學習 21
2.4.1 Q-Learning 和 Deep Q-Network（DQN） 21
2.4.2 Random Exploration 22
3. 研究方法 24
3.1 資料庫 24
3.1.1 初始資料庫 24

3.1.2 生成式模型訓練 31
3.1.3 資料擴增方法 36
3.2 物件偵測模型 41
3.3 圖像分割模型 43
3.4 無使用強化學習的檢測流程 45
3.5 使用強化學習的檢測流程 49
3.5.1 鑑別器 50
3.5.2 強化學習策略和權重更新 51
3.5.3 評估指標 56
4. 實驗結果 57
4.1 實驗環境與實驗設置 57
4.1.1 實驗環境 57
4.1.2 實驗數據 57
4.2 資料擴增實驗結果 58
4.2.1 生成鏽蝕影像圖 58
4.2.2 資料擴增對模型性能的影響 59
4.2.3 擴增技術的多樣性與具體操作 60
4.3 強化學習適應於不同資料集的實驗結果 61
4.3.1 新資料集訓練結果 61
4.3.2 提議方法訓練結果 63
4.3.3 人為介入之訓練結果 65
4.3.4 公開資料集測試結果 69
5. 結論 70
Reference 71

參考文獻

Reference

[1] A. Bochkovskiy, C.-Y. Wang, and H.-Y. M. Liao, “Yolov4: Optimal speed and accuracy of object detection,” in arXiv preprint arXiv:.10934, 2020.
[2] N. Carion, F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko, “End-to-end object detection with transformers,” European Conference on Computer Vision, pp. 213-229, 2020.
[3] I. Goodfellow, J. Pouget-Abadie, M. Mirza, B. Xu, D. Warde-Farley, S. Ozair, A. Courville, and Y. Bengio, “Generative adversarial networks,” Communications of the ACM, vol. 63, no. 11, pp. 139-144, 2020.
[4] K. He, G. Gkioxari, P. Dollar, and R. Girshick, “Mask r-cnn,” Proceedings of the IEEE International Conference on Computer Vision, pp. 2961-2969, 2017.
[5] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 770-778, 2016.
[6] J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Advances in Neural Information Processing Systems, vol. 33, no., pp. 6840-6851, 2020.
[7] ISO, “Paints and varnishes — Evaluation of degradation of coatings — Designation of quantity and size of defects, and of intensity of uniform changes in appearance — Part 3: Assessment of degree of rusting.,” International Organization for Standardization, Geneva, Technical Standard ISO 4628-3:2016, 2016.
[8] A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, and W.-Y. Lo, “Segment anything,” Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 4015-4026, 2023.
[9] T.-Y. Lin, P. Dollar, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature pyramid networks for object detection,” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 2117-2125, 2017.
[10] J. Long, E. Shelhamer, and T. Darrell, “Fully convolutional networks for semantic segmentation,” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 3431-3440, 2015.
[11] V. Mnih, “Playing atari with deep reinforcement learning,” in arXiv preprint arXiv:1312.5602., 2013.
[12] PKLot, “Rust Detection Dataset,” Roboflow, 2023.
[13] A. Radford, “Unsupervised representation learning with deep convolutional generative adversarial networks,” in arXiv preprint arXiv:.06434, 2015.
[14] J. Redmon, “You only look once: Unified, real-time object detection,” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp., 2016.
[15] S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards real-time object detection with region proposal networks,” IEEE Transactions on Pattern Analysis Machine Intelligence, vol. 39, no. 6, pp. 1137-1149, 2016.
[16] R. Rombach, A. Blattmann, D. Lorenz, P. Esser, and B. Ommer, “High-resolution image synthesis with latent diffusion models,” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pp. 10684-10695, 2022.
[17] O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” Medical Image Computing and Computer-Assisted Intervention–MICCAI 2015: 18th International Conference, Munich, Germany, October 5-9, 2015, Proceedings, Part III 18, pp. 234-241, 2015.
[18] R. S. Sutton and A. G. Barto, “Reinforcement learning: an introduction MIT Press,” Cambridge, MA, 22447, 10., vol. 22447, no., pp. 10, 1998.
[19] C. J. Watkins and P. Dayan, “Q-learning,” Machine learning,8, 279-292., vol. 8, no., pp. 279-292, 1992.
[20] L. Zhang, A. Rao, and M. Agrawala, “Adding conditional control to text-to-image diffusion models,” Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 3836-3847, 2023.
[21] J.-Y. Zhu, T. Park, P. Isola, and A. A. Efros, “Unpaired image-to-image translation using cycle-consistent adversarial networks,” Proceedings of the IEEE International Conference on Computer vision, pp. 2223-2232, 2017.

指導教授

林智揚(Chih-Yang Lin)

審核日期

2025-1-15

推文