臺灣博碩士論文加值系統

近年來隨著人工智慧(AI; Artificial Intelligence)的發展，深度學習(DL; Deep Learning)的應用不斷的擴大。不論是在交通、製造、電子、醫學研究等領域都已廣泛應用。為了更精確對任務進行洞察和預測，如何提升模型的性能已成為一種趨勢。然而對於不同的任務，不同的問題也會接踵而來。另外極端氣候變化對道路安全造成越來越多的威脅，因暴雨沖刷、土石流、地震頻繁產生路面坑洞和裂縫，在下雨時路面容易形成不規則的積水，影響交通道路安全。因此可以利用人工智慧對道路進行偵測，透過人工智慧的協助可以對特定區域進行偵測，並且可以減少人力的支出，準確的偵測(Detection)和分割(Segmentation)這些隱患成為保護用路人安全的關鍵。本研究採用U-Net作為改進的基礎模型，該模型在語意分割任務中有顯著的突破，因此提出了一種新型的密集連接U-Net （DCU-Net; Dense Connection U-Net）架構，由於積水的偵測和分割問題的複雜性很高，因為它具備不規則大小及形狀，也可能是清澈或混濁的積水，且都具有反光特性，容易將周遭環境也偵測為積水區域。DCU-Net使用DenseNet121作為編碼器，DenseNet在語意分割及目標檢測上的精準度有顯著的表現，通過特徵重用來減少模型的參數數量，並利用跳躍連接增強編碼階段和解碼階段之間的特徵傳遞，從而有效恢復特徵的細節。與現有的文獻比較時，在考慮實際積水的情況下，本論文提出的方法能夠更準確的進行偵測與分割。在模型評估指標結果中對準確率(Accuracy)、精確率(Precision)、召回率(Recall)、F1-scorec 和交並比(IOU; Intersection over Union)進行比較，不論是哪種評估指標，本論文提出的方法皆優於其他現有文獻。

In recent years, with the development of artificial intelligence (AI), the applications of deep learning have been continuously expanding. It has been widely applied in various fields such as transportation, manufacturing, electronics, and medical research. To gain more precise insights and predictions for specific tasks, improving model performance has become a trend. However, different tasks bring different challenges. Additionally, extreme climate changes have posed increasing threats to road safety. Heavy rains, landslides, and frequent earthquakes create potholes and cracks on road surfaces, making roads prone to irregular water accumulation during rain, which affects traffic safety. Therefore, AI can be used for road detection. With the assistance of AI, specific areas can be monitored which help in reducing manpower expenditure. Accurate detection and segmentation of these hazards are crucial for protecting road users' safety. This study uses U-Net as the base model for improvement, as it has shown significant breakthroughs in semantic segmentation tasks. Hence, a novel Dense Connected U-Net (DCU-Net) architecture is proposed. Due to the high complexity of water detection and segmentation problems, which involve irregular sizes and shapes, and the water can be clear or turbid with reflective properties that often mislead the detection of surrounding environments as water areas, DCU-Net adopts DenseNet121 as the encoder. DenseNet has shown significant performance in semantic segmentation and object detection. It reduces the number of model parameters through feature reuse and enhances feature transmission between the encoding and decoding stages using skip connections, effectively recovering feature details. Compared with existing literature, and considering actual water accumulation scenarios, the method proposed in this paper can achieve more accurate detection and segmentation. In the model evaluation metrics, including accuracy, precision, recall, F1-score and IOU the proposed method outperforms other existing literature in all evaluation metrics.

摘要...i
Abstract...ii
誌謝...iv
目錄...v
表目錄...viii
圖目錄...x
第一章 緒論...1
1.1 研究背景與目的...1
1.2 論文架構...3
第二章 文獻綜述...4
2.1 圖像分割相關研究...4
2.2 卷積神經網路(Convolutional Neural Networks)...8
2.2.1 卷積層(Convolution Layer)...8
2.2.2 深度(Depth)...10
2.2.3 步長(Stride)...11
2.2.4 填充(Padding)...12
2.2.5 反卷積層(Deconvolution Layer)...13
2.2.6 池化層(Pooling Layer)...14
2.2.7 激勵函數(Activation Function)...16
2.2.8 批次正規化(Batch Normalization)...18
2.3 語意分割(Semantic Segmentation)...20
2.3.1 U-Net...21
2.3.2 DeepLabv3...23
2.3.3 DeepLabv3+...24
2.3.4 PSPNet...25
2.3.5 ResU-Net...26
2.3.6 DenseNet...27
2.4 最佳化器(Optimizer)...28
2.4.1 隨機梯度下降(Stochastic Gradient Descent)...29
2.4.2 自適應矩估計(Adaptive Moment Estimation)...29
2.4.3 修正的自適應矩估計(Rectified Adaptive Moment Estimation)...31
第三章 研究方法與內容...32
3.1 改進的模型架構...32
3.2 模型架構描述...33
3.3 損失函數(Loss Function)...36
3.4 評估指標...37
3.5 模型訓練流程...39
3.6 資料集介紹...40
3.6.1 資料增強(Data Augmentation)...40
3.6.2 圖像標註(Image Annotation)...42
第四章 最佳化器與學習率的比較...43
4.1 實驗環境...43
4.2 基於不同模型使用不同最佳化器...43
4.2.1 實驗結果與分析...45
4.3 基於不同模型使用不同學習率...50
4.3.1 實驗結果與分析...51
4.4 討論與結論...60
第五章 基於資料預處理的比較...61
5.1 資料預處理(Data Preprocessing)...61
5.2 實驗結果與分析...65
5.3 討論與結論...78
第六章 使用注意力機制的比較...79
6.1 混合注意力...79
6.1.1 通道注意力模組...80
6.1.2 空間注意力模組...83
6.2 基於混合注意力的模型架構...85
6.3 實驗結果與分析...87
6.4 討論與結論...94
第七章 結論...95
參考文獻...97
Extended Abstract...102

[1] K. He, X. Zhang, S. Ren and J. Sun, "Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 37, no. 9, pp. 1904-1916, 1 Sept. 2015.
[2] H. -C. Shin et al., "Deep Convolutional Neural Networks for Computer-Aided Detection: CNN Architectures, Dataset Characteristics and Transfer Learning," IEEE Transactions on Medical Imaging, vol. 35, no. 5, pp. 1285-1298, May 2016.
[3] T. He, W. Huang, Y. Qiao and J. Yao, "Text-Attentional Convolutional Neural Network for Scene Text Detection," IEEE Transactions on Image Processing, vol. 25, no. 6, pp. 2529-2541, June 2016.
[4] R. Girshick, J. Donahue, T. Darrell and J. Malik, "Region-Based Convolutional Networks for Accurate Object Detection and Segmentation," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 38, no. 1, pp. 142-158, 1 Jan. 2016.
[5] F. Fuentes-Peñailillo, S. Ortega-Farías, M. Rivera, M. Bardeen and M. Moreno, "Using clustering algorithms to segment UAV-based RGB images," 2018 IEEE International Conference on Automation/XXIII Congress of the Chilean Association of Automatic Control (ICA-ACCA), Concepcion, Chile, 2018, pp. 1-5.
[6] P. Mylvaganam and M. B. Dissanayake, "Deep Learning for Arbitrary-Shaped Water Pooling Region Detection on Aerial Images," 2022 Moratuwa Engineering Research Conference (MERCon), Moratuwa, Sri Lanka, 2022, pp. 1-5.
[7] J. Fu, J. Liu, H. Tian, Y. Li, Y. Bao, Z. Fang and H. Lu, "Dual Attention Network for Scene Segmentation," 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Long Beach, CA, USA, 2019, pp. 3141-3149.
[8] Q. Wang, B. Wu, P. Zhu, P. Li, W. Zuo and Q. Hu, "ECA-Net: Efficient Channel Attention for Deep Convolutional Neural Networks," 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Seattle, WA, USA, 2020, pp. 11531-11539.
[9] S. V. Fani, Kamirul, A. Noer and S. Y. C. Bissa, "U-Net Based Water Region Segmentation for LAPAN-A2 MSI," 2022 IEEE International Conference on Aerospace Electronics and Remote Sensing Technology (ICARES), Yogyakarta, Indonesia, 2022, pp. 1-5.
[10] J. -J. Qiao, X. Wu, J. -Y. He, W. Li and Q. Peng, "SWNet: A Deep Learning Based Approach for Splashed Water Detection on Road," IEEE Transactions on Intelligent Transportation Systems, vol. 23, no. 4, pp. 3012-3025, April 2022.
[11] M. G. Prasad, A. Chakraborty, R. Chalasani and S. Chandran, "Quadcopter-based Stagnant Water Identification," 2015 Fifth National Conference on Computer Vision, Pattern Recognition, Image Processing and Graphics (NCVPRIPG), Patna, India, 2015, pp. 1-4.
[12] X. Zhang and S. Zhao, "Semantic Segmentation of Water Body in High-Resolution Remote Sensing Images Based on DeepLabV3+," 2023 3rd International Conference on Electronic Information Engineering and Computer Science (EIECS), Changchun, China, 2023, pp. 1061-1064.
[13] H. -Y. Lin, S. -L. Tseng and J. -Y. Li, "SUR-Net: A Deep Network for Fish Detection and Segmentation With Limited Training Data," IEEE Sensors Journal, vol. 22, no. 18, pp. 18035-18044, 15 Sept.15, 2022.
[14] D. Qiu, M. Xiao, S. Wan, C. Qin and Z. Zhu, "Pavement Crack Detection in Infrared Images Using a DCNN and CCL Algorithm," IEEE Sensors Journal, vol. 23, no. 5, pp. 4548-4555, 1 March1, 2023.
[15] L. Wang and H. Wang, "Water Hazard Detection Using Conditional Generative
Adversarial Network With Mixture Reflection Attention Units," IEEE Access, vol. 7, pp. 167497-167506, 2019.
[16] H. Samaranayake, O. Mudannayake, D. Perera, P. Kumarasinghe, C. Suduwella, K. De Zoysa and P. Wimalaratne, (2023). Detecting Water in Visual Image Streams from UAV With Flight Constraints. Journal of Visual Communication and Image Representation, 96, 103933.
[17] W. Wang and C. Su, "Convolutional Neural Network-Based Pavement Crack Segmentation Using Pyramid Attention Network," IEEE Access, vol. 8, pp. 206548-206558, 2020.
[18] B. Wang, Z. Chen, L. Wu, X. Yang and Y. Zhou, "SADA-Net: A Shape Feature Optimization and Multiscale Context Information-Based Water Body Extraction Method for High-Resolution Remote Sensing Images," IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 15, pp. 1744-1759, 2022.
[19] Q. Cai, R. Wan, H. Li, C. Wang and H. Chang, "Remote Sensing Image River Segmentation Method Based on U-Net," 2022 IEEE 8th International Conference on Cloud Computing and Intelligent Systems (CCIS), Chengdu, China, 2022, pp. 215-220.
[20] P. Nie, X. Cheng, Z. Song, M. Mao, T. Wang and L. Meng, "Rethinking BiSeNet: A Lightweight Network for Urban Water Extraction," IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1-10, 2023.
[21] T. Balaji, V. Khanna and T. Nalini, "Deep Learning Technique For Effective Segmentation Of Water Bodies Using Remote Sensing Images," 2023 5th International Conference on Inventive Research in Computing Applications (ICIRCA), Coimbatore, India, 2023, pp. 157-160.
[22] Z. Li, R. Wang, W. Zhang, F. Hu and L. Meng, "Multiscale Features Supported DeepLabV3+ Optimization Scheme for Accurate Water Semantic Segmentation," IEEE Access, vol. 7, pp. 155787-155804, 2019.
[23] W. Feng, H. Sui, W. Huang, C. Xu and K. An, "Water Body Extraction From Very High-Resolution Remote Sensing Imagery Using Deep U-Net and a Superpixel-Based Conditional Random Field Model," IEEE Geoscience and Remote Sensing Letters, vol. 16, no. 4, pp. 618-622, April 2019.
[24] K. Yuan, X. Zhuang, G. Schaefer, J. Feng, L. Guan and H. Fang, "Deep-Learning-Based Multispectral Satellite Image Segmentation for Water Body Detection," IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 14, pp. 7422-7434, 2021.
[25] X. Zhang, J. Li and Z. Hua, "MRSE-Net: Multiscale Residuals and SE-Attention Network for Water Body Segmentation From Satellite Images," IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 15, pp. 5049-5064, 2022.
[26] H. -F. Zhong, Q. Sun, H. -M. Sun and R. -S. Jia, "NT-Net: A Semantic Segmentation Network for Extracting Lake Water Bodies From Optical Remote Sensing Images Based on Transformer," IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 113, 2022.
[27] M. Talal, A. Panthakkan, H. Mukhtar, W. Mansoor, S. Almansoori and H. A. Ahmad, "Detection of Water-Bodies Using Semantic Segmentation," 2018 International Conference on Signal Processing and Information Security (ICSPIS), DUBAI, United Arab Emirates, 2018, pp. 1-4.
[28] S. Dong, L. Pang, Y. Zhuang, W. Liu, Z. Yang and T. Long, "Optical Remote Sensing Water-Land Segmentation Representation Based on Proposed SNS-CNN Network," IGARSS 2019 - 2019 IEEE International Geoscience and Remote Sensing Symposium, Yokohama, Japan, 2019, pp. 3895-3898.
[29] R. Xin, J. Zhang and Y. Shao, "Complex Network Classification With Convolutional Neural Network," Tsinghua Science and Technology, vol. 25, no. 4, pp. 447-457, Aug. 2020.
[30] S. Ioffe and C. Szegedy, “Batch Normalization: Accelerating Deep Network Training By Reducing Internal Covariate Shift,” in International conference on machine learning, pp. 448–456, PMLR, 2015.
[31] O. Ronneberger, P. Fischer, and T. Brox. “U-Net: Convolutional Networks For Biomedical Image Segmentation,” in Proc. Int. Conf. Med.Image Comput.-Assist. Intervent. Cham, Switzerland: Springer, 2015, pp. 234–241.
[32] L. C. Chen, G. Papandreou, F. Schroff, and H. Adam. “Rethinking Atrous Convolution For Semantic Image Segmentation.” arXiv preprint arXiv:1706.05587 (2017).
[33] L. C. Chen, G. Papandreou, F. Schroff, and H. Adam. “Encoder-decoder With Atrous Separable Convolution For Semantic Image Segmentation.” Proceedings of the European conference on computer vision (ECCV), 2018, pp. 801-818.
[34] H. Zhao, J. Shi, X. Qi, X. Wang and J. Jia. “Pyramid Scene Parsing Network.” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 2017, pp. 2881-2890.
[35] G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, “Densely Connected Convolutional Networks,” in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), Jul. 2017, pp. 4700–4708.
[36] S. Ruder, ‘‘An Overview Of Gradient Descent Optimization Algorithms,’’ 2016, arXiv:1609.04747.
[37] D. P. Kingma and J. Ba, ‘‘Adam: A Method For Stochastic Optimization’’, 2014, arXiv:1412.6980.
[38] L. Liu, H. Jiang, P. He, W. Chen, X. Liu, J. Gao, and J. Han, ‘‘On The Variance Of The Adaptive Learning Rate And Beyond’’, 2019, arXiv:1908.03265.
[39] A. Mao, M. Mohri, Y. Zhong. “Cross-entropy Loss Functions: Theoretical Analysis And Applications.” International Conference on Machine Learning, PMLR, pp. 23803-23828, 2023. arXiv preprint arXiv:2304.07288.
[40] Stagnant Water, “Stagnant Water Dataset,” Roboflow, 2024. [Online]. Available: https://universe.roboflow.com/stagnant-water/stagnant-water.[Accessed: 22-Apr-2024].
[41] S. Woo, J. Park, J. Y. Lee, and I. S. Kweon, “Cbam: Convolutional Block Attention Module.” Proceedings of the European conference on computer vision (ECCV), 2018. pp. 3-19.
[42] J. -H. Han, S. Yang and B. -U. Lee, "A Novel 3-D Color Histogram Equalization Method With Uniform 1-D Gray Scale Histogram," in IEEE Transactions on Image Processing, vol. 20, no. 2, pp. 506-512, Feb. 2011.
[43] D. Wei, P. Li, H. Xie and Y. Xu, "DRCNet: Road Extraction From Remote Sensing Images Using DenseNet With Recurrent Criss-Cross Attention and Convolutional Block Attention Module," in IEEE Access, vol. 11, pp. 126879-126891, 2023.