跳到主要內容

臺灣博碩士論文加值系統

(44.211.31.134) 您好!臺灣時間:2024/07/25 18:14
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果 :::

詳目顯示

我願授權國圖
: 
twitterline
研究生:蘇永騰
研究生(外文):SU, YUNG-TENG
論文名稱:基於兩階段特徵融合及注意力機制之語義分割網路
論文名稱(外文):Semantic Segmentation Network based on Two-Stage Feature Fusion and Attention Mechanism
指導教授:許宏銘許宏銘引用關係
指導教授(外文):Michael Nobert Mayer
口試委員:徐超明˙莊貴枝
口試日期:2023-07-13
學位類別:碩士
校院名稱:國立中正大學
系所名稱:電機工程研究所
學門:工程學門
學類:電資工程學類
論文種類:學術論文
論文出版年:2023
畢業學年度:111
語文別:英文
論文頁數:49
中文關鍵詞:電腦視覺注意力機制特徵融合語義分割影像切割
外文關鍵詞:Computer visionimage segmentationsemantic segmentationfeature fusionattention mechanism
相關次數:
  • 被引用被引用:0
  • 點閱點閱:107
  • 評分評分:
  • 下載下載:0
  • 收藏至我的研究室書目清單書目收藏:0
本文提出了基於深度學習的語義分割方法,使用兩種不同的卷積神經網路以進行特徵提取並對其進行融合特徵,最後使用全卷積網路[1]來對融合後的特徵圖進行像素級的分類。使用AlexNet[2],VGG-16[3]以及其預訓練權重提取影像的特徵,並將特徵傳入全卷積網路,使用轉置卷積生成與原始影像同大小的輸出。為了使特徵能夠融合,使用全局平均池化以及1×1卷積核的卷積層。並使用通道注意力機制[4]提高特徵的表達能力。本文使用Pascal數據集來進行實驗並與全卷積網路進行比較,證明了兩階段特徵融合及注意力機制在影像切割能提高模型效能。
This paper proposes a deep learning-based semantic segmentation method that utilizes two different convolutional neural networks (CNNs) for feature extraction and fuses the extracted features. The fused feature map is then passed through a fully convolutional network (FCN) [1] for pixel-level classification. AlexNet [2] and VGG-16 [3], along with their pre-trained weights, are employed to extract features from the images, which are subsequently fed into the FCNs. Transposed convolutions are utilized to generate output maps of the same size as the original image. To enable feature fusion, global average pooling, and 1×1 convolutional layers are employed. Furthermore, a channel attention mechanism [4] is incorporated to enhance the expressive power of the features. The Pascal dataset is used for experimental evaluation, and a comparison is made against the FCN model to demonstrate the effectiveness of the two-stage feature fusion and attention mechanism in improving model performance in image segmentation tasks.
Acknowledgment i
中文摘要 ii
abstract iii
Table of Contents iv
List of Figures vi
List of Tables vii
List of Equations viii
List of Varaible ix
1. Introduction 1
2. Related Work 3
2.1. Artificial Neural Network 3
2.2. Convolution Neural Network 4
2.2.1. Convolution Layer 4
2.2.2. Pooling Layer 5
2.3. Semantic Segmentation 7
2.4. Fully Convolution Network 8
2.4.1. Initialization with VGG-16 8
2.4.2. Transposed Convolution 8
2.4.3. Matrix Degradation 9
2.4.4. Skip Connection 10
2.4.5. Bilinear Interpolation 11
2.5. Two Stage Feature Fusion 13
2.5.1. Multi-scale Feature Fusion 13
2.5.2. Two Stage Feature Fusion 13
2.6. Channel Attention 16
3. Implementation 18
3.1. FCNs 18
3.2. Two Stage Feature Fusion 20
3.2.1. VGG-16 22
3.2.2. AlexNet 23
3.3. Channel Attention 25
4. Experimental Results 27
4.1. Critic 27
4.2. Result 27
4.2.1. Data Result 28
4.2.2. Stability Result 29
4.2.3. Image Result 32
5. Summary and Conclusions 35
6. Reference 36
Appendix 38

[1]Long, J., Shelhamer, E., & Darrell, T. (2015). Fully convolutional networks for semantic segmentation. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 3431-3440).
[2]A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet Classification with Deep Convolutional Neural Networks,” in Advances in Neural Information Processing Systems, Curran Associates, Inc., 2012. Accessed: Jun. 05, 2023. [Online]. Available: https://proceedings.neurips.cc/paper_files/paper/2012/hash/c399862d3b9d6b76c8436e924a68c45b-Abstract.html
[3]K. Simonyan and A. Zisserman, “Very Deep Convolutional Networks for Large-Scale Image Recognition.” arXiv, Apr. 10, 2015. doi: 10.48550/arXiv.1409.1556.
[4]J. Fan et al., “Multi-Scale Feature Fusion: Learning Better Semantic Segmentation for Road Pothole Detection,” in 2021 IEEE International Conference on Autonomous Systems (ICAS), Aug. 2021, pp. 1–5. doi: 10.1109/ICAS49788.2021.9551165.
[5]K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition.” arXiv, Dec. 10, 2015. doi: 10.48550/arXiv.1512.03385.
[6]M. Lin, Q. Chen, and S. Yan, “Network In Network.” arXiv, Mar. 04, 2014. doi: 10.48550/arXiv.1312.4400.
[7]O. Ronneberger, P. Fischer, and T. Brox, “U-Net: Convolutional Networks for Biomedical Image Segmentation.” arXiv, May 18, 2015. doi: 10.48550/arXiv.1505.04597.
[8]L.-C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. Yuille, “DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs.” arXiv, May 11, 2017. doi: 10.48550/arXiv.1606.00915.
[9]“物體分割與物件識別模型概述 (2).” https://linchin.ndmctsgh.edu.tw/Deep%20Learning-Theory%20and%20Practice/Lesson%2011/Lesson_11.html#(2) (accessed Jun. 28, 2023).
[10]A. E. Orhan and X. Pitkow, “Skip Connections Eliminate Singularities.” arXiv, Mar. 04, 2018. doi: 10.48550/arXiv.1701.09175.
[11]E. Orhan, “Why is it hard to train deep neural networks? Degeneracy, not vanishing gradients, is the key,” Severely Theoretical, Jan. 01, 2018. https://severelytheoretical.wordpress.com/2018/01/01/why-is-it-hard-to-train-deep-neural-networks-degeneracy-not-vanishing-gradients-is-the-key/ (accessed May 25, 2023).
[12]Lin, T. Y., Dollár, P., Girshick, R., He, K., Hariharan, B., & Belongie, S. (2017). Feature pyramid networks for object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 2117-2125).
[13]Huang, G., Liu, Z., Van Der Maaten, L., & Weinberger, K. Q. (2017). Densely connected convolutional networks. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 4700-4708).
[14]Y. Liu, Y. Liu, and L. Ding, “Scene Classification Based on Two-Stage Deep Feature Fusion,” IEEE Geoscience and Remote Sensing Letters, vol. 15, no. 2, pp. 183–186, Feb. 2018, doi: 10.1109/LGRS.2017.2779469.
[15]“从全连接层的变化看懂卷积神经网络的奥秘_阿猫的自拍的博客-CSDN博客.” https://blog.csdn.net/weixin_37721058/article/details/102479612 (accessed Jun. 30, 2023).
[16]A. Vaswani et al., “Attention Is All You Need.” arXiv, Dec. 05, 2017. Accessed: Jun. 19, 2023. [Online]. Available: http://arxiv.org/abs/1706.03762

電子全文 電子全文(網際網路公開日期:20280725)
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top
無相關期刊