生成協作網路(Generative Collaborative Network, GCN)是一用來輔助醫生進行大腸內視鏡檢查的神經網路，基於其輕巧的架構特性，作者本先欲以其為核心架構進行應用系統開發，然而我們發現其採用影像處理方式進行偵測框的繪製時，常產生雜框導致系統效能降低。
本論文提出一以生成協作網路為主體的檢測系統，在原網路新增兩個網路來改善既有問題並提升效能，分別是區域提議網路(Region Proposal Network, RPN)及注意力機制介面(Hard Attention Interface, HAI)，區域提議網路是利用特徵圖的資訊進行位置偵測，因此我們將生成協作網路生成的預測區域作為病灶的概略位置，再提供生成網路產生的特徵圖使區域提議網路在概略位置對應的特徵圖區域上進行位置的偵測，如此一來我們便能得到更精準的標記區域同時避免掉在其他地方產生雜框。緊接著，我們將偵測出位置的原始大腸鏡影像送入注意力機制介面，利用其經由多次特徵擷取所累積的資訊進行病灶的種類分析。
本論文所使用的大腸鏡資料除了來自CVC-ClinicDB以及CVC-EndoSceneStill等兩個個資料庫外，也包含來自合作醫院的臨床病例資料，我們將利用上述資料進行實驗，證明本論文所提出之系統確實能有效解決開頭所點出的問題，證明本系統的偵測表現更優於前者。

Generative Collaborative Network (GCN) is a dedicated neural network proposed to support the automatic diagnosis for the colonoscopy. However, while trying to develop a diagnosis system, we found it uses the detected image to form the bounding boxes that always produces many unnecessary bounding boxes. It decreases the efficiency of the location and recognition. Therefore, we equip GCN with two new additional networks, region proposal network (RPN) for target pinpoint and hard attention interface (HAI) for classification enhancement. RPN can adjust the location of bounding boxes according to the information of feature maps. Therefore, we take the prediction from GCN as a target candidate. Then RPN focuses mainly on this region’s feature map. In this way, we not only get a more precise detection but also avoid unnecessary bounding boxes. And HAI classifies the detection from RPN. The data used in this thesis comes from CVC-ClinicDB, CVC-EndoSceneStill datasets, and the collaborative hospital. We present the experiment results to prove that the proposed system can get rid of the aforementioned problem and outperform the original GCN.

論文審定書 i
中文摘要 iii
Abstract iv
目錄 v
圖目錄 vii
表目錄 ix
第1章 緒論 1
1.1研究動機 1
1.2文獻回顧 2
1.3論文架構 2
第2章 研究背景 3
2.1卷積神經網路(Convolution Neural Network) 3
2.1.1卷積層(Convolution Layer) 4
2.1.2池化層(Pooling Layer) 5
2.1.3全連接層(Fully Connected Layer) 6
2.2生成協作神經網路(Generative Collaborative Network) 7
2.2.1網路流程 7
2.2.2網路架構 8
2.3 Faster RCNN 10
2.3.1網路架構 10
2.4 Recurrent Attention Model 12
2.4.1模型架構 12
第3章 研究方法 14
3.1核心網路 14
3.1.1生成協作網路 14
3.1.2區域提議網路 16
3.1.3注意力介面 18
3.2擴增生成協作網路(Extended-GCN) 19
3.2.1定位 20
3.2.2分類 22
3.3擴增生成協作網路訓練流程 23
第4章 實驗 34
4.1實驗環境與樣本介紹 34
4.2實驗說明 36
4.3實驗結果展示 38
4.3.1定位實驗 38
4.3.2分類實驗 42
第5章 結論與未來展望 51
5.1結論 51
5.2未來展望 52
參考文獻 53

[1] M. Hwang et al., "An Adaptive Regularization Approach to Colonoscopic Polyp Detection Using a Cascaded Structure of Encoder–Decoders," International Journal of Fuzzy Systems, vol. 21, no. 7, pp. 2091-2101, 2019.
[2] G. E. Hinton and R. S. Zemel, "Autoencoders, Minimum Description Length And Helmholtz Free Energy," in Advances in Neural Information Processing Systems, 1994, pp. 3-10.
[3] I. Goodfellow et al., "Generative Adversarial Nets," in Advances in Neural Information Processing Systems, 2014, pp. 2672-2680.
[4] 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.
[5] P. Vincent, H. Larochelle, I. Lajoie, Y. Bengio, P.-A. Manzagol, and L. Bottou, "Stacked Denoising Autoencoders: Learning Useful Representations in A Deep Network with a Local Denoising Criterion," Journal of Machine Learning Research, vol. 11, no. 12, pp.3371-3408, 2010.
[6] S. Ren, K. He, R. Girshick, and J. Sun, "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks," in Advances in Neural Information Processing Systems, pp. 91-99, 2015.
[7] R. Girshick, "Fast R-CNN," in Proceedings of the IEEE International Conference on Computer Vision, pp. 1440-1448, 2015.
[8] K. Simonyan and A. Zisserman, "Very Deep Convolutional Networks for Large-Scale Image Recognition," in ICRL, 2015.
[9] J. Long, E. Shelhamer, and T. Darrell, "Fully Convolutional Networks for Semantic Segmentation," in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 3431-3440, 2015.
[10] V. Mnih, N. Heess, and A. Graves, "Recurrent Models of Visual Attention," in Advances in Neural Information Processing Systems, pp. 2204-2212, 2014.
[11] V. H. Phung and E. J. Rhee, "A High-Accuracy Model Average Ensemble of Convolutional Neural Networks for Classification of Cloud Image Patches on Small Datasets," Applied Sciences, vol. 9, no. 21, p. 4500, 2019.
[12] Y.-L. Boureau, J. Ponce, and Y. LeCun, "A Theoretical Analysis of Feature Pooling in Visual Recognition," in Proceedings of The 27th International Conference on Machine Learning (ICML-10), pp. 111-118, 2010.
[13] J. Zhao, M. Mathieu, and Y. LeCun, "Energy-Based Generative Adversarial Network," Arxiv Preprint Arxiv:1609.03126, 2016.
[14] S. Khan, H. Rahmani, S. A. A. Shah, and M. Bennamoun, "A Guide to Convolutional Neural Networks for Computer Vision," Synthesis Lectures on Computer Vision, vol. 8, no. 1, pp. 1-207, 2018.
[15] 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, pp. 580-587, 2014.
[16] K. He, X. Zhang, S. Ren, and J. Sun, "Deep Residual Learning for Image Recognition," in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 770-778, 2016.
[17] S. Hochreiter and J. Schmidhuber, "Long Short-Term Memory," Neural Computation, vol. 9, no. 8, pp. 1735-1780, 1997.
[18] R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction. MIT press, 2018.
[19] G. Bradski and A. Kaehler, Learning OpenCV: Computer Vision with the Opencv Library. O'Reilly Media, Inc. , 2008.
[20] C. M. Bishop, Pattern Recognition and Machine Learning. Springer, 2006.
[21] R. S. Sutton, D. A. McAllester, S. P. Singh, and Y. Mansour, "Policy Gradient Methods for Reinforcement Learning with Function Approximation," in Advances in Neural Information Processing Systems, pp. 1057-1063, 2000.
[22] G. Fernandez-Esparrach et al., "Exploring the Clinical Potential of an Automatic Colonic Polyp Detection Method Based on the Creation of Energy Maps," Endoscopy, vol. 48, no. 09, pp. 837-842, 2016.
[23] J. Bernal, F. J. Sánchez, G. Fernández-Esparrach, D. Gil, C. Rodríguez, and F. Vilariño, "WM-DOVA Maps for Accurate Polyp Highlighting in Colonoscopy: Validation vs. Saliency Maps from Physicians," Computerized Medical Imaging and Graphics, vol. 43, pp. 99-111, 2015.
[24] D. Vázquez et al., "A Benchmark for Endoluminal Scene Segmentation of Colonoscopy Images," Journal of Healthcare Engineering, vol. 2017, 2017.
[25] S. Arlot and A. Celisse, "A Survey of Cross-Validation Procedures for Model Selection," Statistics Surveys, vol. 4, pp. 40-79, 2010.
[26] P. Goyal et al., "Accurate, Large Minibatch Sgd: Training Imagenet in 1 Hour," arXiv preprint arXiv:1706.02677, 2017.