高分辨遥感图像分类是遥感图像分析解译中一个既实用又关键的步骤，其目的在于依据每幅图像所包含的特征信息确定其所属的语义类别，能够揭示图像的地物结构和本质，为遥感图像目标识别与检测奠定了基础，在自然灾害预防、城市规划、环境监测等领域中的作用也日渐突出。因此，本文以高分辨遥感图像为研究对象，以深度学习为研究基础，针对深度特征难以解译、单一深度特征鲁棒性较差、卷积神经网络（Convolutional Neural Network，CNN）缺乏多尺度细节纹理特征感受野等问题，展开了基于深度纹理特征融合的高分辨遥感图像分类算法研究。本文的主要研究工作概括如下：

（1）针对CNN提取的深度特征缺乏多尺度纹理特征感受野的问题，本文提出了Gabor卷积神经网络（Gabor Convolutional Neural Network，Gabor-CNN），实现了深度Gabor特征的有效提取。该网络选择具有较少参数量的DenseNet-201作为基准模型，并在此基础上将部分标准卷积核替换为二维Gabor核来提取图像中的多尺度、多方向纹理特征。在NWPU-RESISC45实测数据集分类实验中取得了94.08%的准确率，证明了深度Gabor特征网络在遥感图像场景分类中的有效性。

（2）针对CNN提取的深度特征关联性不强、特征图冗余较多等问题，本文提出了小波卷积神经网络（Wavelet Convolutional Neural Network，Wave-CNN），实现了深度小波特征的有效提取。该模型用小波变换分解的低频分量替代传统池化层，以提取更为丰富的全局特征，并结合空间注意力机制，通过高频分量为图像的纹理位置赋予更高的权重，将小波低频分量作为空间注意力模块的输入部分，与基于高频分量的空间权重加权得到空间注意力图。然后，使用通道注意力模块进一步强化图像的特征表达。最后，使用级联融合强化多级小波特征复用。在AID30实测数据集分类实验中取得了96.77%的准确率，证明了结合小波变换和注意力机制能够有效提高高分辨遥感图像场景分类的准确性。

（3）针对单一深度纹理特征模型的分类性能存在瓶颈，鲁棒性较差的问题，本文提出了基于Dempster-Shafer（D-S）证据理论的深度纹理特征融合网络（Hierarchical Deep Texture Features Fusion Network，HDTFF-Net）。该网络通过决策级融合，将DenseNet-201、Gabor-CNN和Wave-CNN三种层次特征集成于一个网络，进而实现来自不同独立源深度特征集的有效融合。在NWPU-RESISC45、AID30和PatternNet38实测数据集分类实验中分别取得了94.47%、97.46%和99.64%的准确率，证实了所提出的HDTFF-Net网络在高分辨遥感图像场景分类中的有效性和可行性，相比于其他方法，HDTFF-Net具有更高的分类性能和更好的鲁棒性。

High-resolution remote sensing image classification is a crucial and practical step in the analysis and interpretation of remote sensing images. It aims to determine the semantic category of each image according to its features. This process can well reveal the structure and nature in image, thus laying a solid foundation for remote sensing image target identification and detection. And it plays an increasingly prominent role in natural disaster prevention, urban planning, environmental monitoring and others. Therefore, for high-resolution remote sensing images, the classification on the basis of deep texture feature fusion for addressing the challenges, including the difficulty in interpreting deep features, poor robustness of single deep features, and the lack of multi-scale detailed texture feature receptive fields in Convolutional Neural Networks (CNN). The main contents of this dissertation are summarized as follows.

(1) Focusing on the lack of multi-scale texture feature receptive fields in deep features extracted by CNN, this thesis proposes the Gabor Convolutional Neural Network (Gabor-CNN) for effectively extracting the deep Gabor features from high-resolution remote sensing images. The Gabor-CNN employs the DenseNet-201 as the baseline model considering its fewer parameters. Based on this model, some standard convolutional kernels are replaced with two-dimensional Gabor kernels to extract multi-scale and multi-directional texture features from the images. In the classification experiment on the NWPU-RESISC45 dataset, the proposed Gabor-CNN achieves an accuracy of 94.08%, which well demonstrates the effectiveness of Gabor-CNN in remote sensing image scene classification.

(2) To address the weak correlation among deep features extracted by CNNs and the excessive redundancy in feature maps, this thesis proposes the Wavelet Convolutional Neural Network (Wave-CNN) for effectively extracting deep wavelet features. In Wave-CNN, traditional pooling layers are replaced by the low-frequency components obtained by wavelet transform for extracting richer global features. Additionally, a spatial attention mechanism is incorporated, which assigns higher weights to the texture positions of the image through the high-frequency components. The low-frequency components by wavelet transform serve as the inputs of the spatial attention module, and then the spatial attention maps are obtained by the weighted fusion of the inputs and the weights of high-frequency components. Finally, a channel attention module is constructed for further enhancing the feature representation of the image. And a cascade fusion technique is used to reinforce the reuse of multi-level wavelet features. The classification experiments on the AID30 dataset shows that the proposed Wave-CNN achieved an accuracy of 96.77%, demonstrating that the combination of wavelet transform and attention mechanisms can effectively improve the performance of high-resolution remote sensing image scene classification.

(3) For addressing the limited performance and the poor robustness of single deep texture feature models in classification, this thesis proposes the Hierarchical Deep Texture Features Fusion Network (HDTFF-Net). On the basis of the Dempster-Shafer (D-S) evidence theory, the proposed  HDTFF-Net integrates three hierarchical sub-networks, namely DenseNet-201, Gabor-CNN, and Wave-CNN, through a decision-level fusion. In this way, the HDTFF-Net can effectively fuse the deep features from different independent networks. Experiments on the NWPU-RESISC45, AID30, and PatternNet38 datasets shows that the HDTFF-Net achieves the accuracies of 94.47%, 97.46%, and 99.64%, respectively, thus verifying the effectiveness and feasibility of the proposed HDTFF-Net in high-resolution remote sensing image scene classification. Compared to other methods, HDTFF-Net exhibits better classification performance and greater robustness.

[1] 沈金悦. 基于卷积神经网络和注意力机制的高光谱遥感图像分类研究[D]. 青岛科技大学, 2022.

[2] 周培诚, 程塨, 姚西文, 韩军伟. 高分辨率遥感影像解译中的机器学习范式[J]. 遥感学报, 2021, 25(1): 182-197.

[3] Aptoula E. Remote sensing image retrieval with global morphological texture descriptors[J]. IEEE Transactions on Geoscience and Remote Sensing, 2013, 52(5): 3023-3034.

[4] Qin G, Qin G. Virtual reality video image classification based on texture features[J]. Complexity, 2021, 2021: 1-11.

[5] Haralick R M, Shanmugam K, Dinstein I H. Textural features for image classification[J]. IEEE Transactions on Systems, Man, and Cybernetics, 1973, 6: 610-621.

[6] Jain A K, Ratha N K, Lakshmanan S. Object detection using Gabor filters[J]. Pattern Recognition, 1997, 30(2): 295-309.

[7] Ojala T, Harwood I. A comparative study of texture measures with classification based on feature distributions[J]. Pattern Recognition, 1996, 29(1): 51-59.

[8] 冯逍, 肖鹏峰, 李琦,等. 三维Gabor滤波器与支持向量机的高光谱遥感图像分类[J]. 光谱学与光谱分析, 2014, 34(8): 2218-2224.

[9] Tan Q, Liu Y, Chen X, et al. Multi-label classification based on low rank representation for image annotation[J]. Remote Sensing, 2017, 9(2): 109.

[10] Chen S B, Wei Q S, Wang W Z, et al. Remote sensing scene classification via multi-branch local attention network[J]. IEEE Transactions on Image Processing, 2021, 31: 99-109.

[11] 廖敏, 陈国郑. 基于灰度共生矩阵的遥感图像分类研究[J]. 黑龙江科技信息, 2018, 28: 20-21.

[12] Kobayashi T. Dirichlet-based histogram feature transform for image classification[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2014: 3278-3285.

[13] 李祥霞, 吉晓慧, 李彬. 细粒度图像分类的深度学习方法[J]. 计算机科学与探索, 2021, 15(10): 1830-1842.

[14] Lecun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324.

[15] Krizhevsky A, Sutskever I, Hinton G E. Imagenet classification with deep convolutional neural networks[C]. Advances in Neural Information Processing Systems, 2012: 1097-1105.

[16] Szegedy C, Liu W, Jia Y, et al. Going deeper with convolutions[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 1-9.

[17] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[J]. ArXiv Preprint ArXiv:1409.1556, 2014: 1-14.

[18] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770-778.

[19] Huang G, Liu Z, Van Der Maaten L, et al. Densely connected convolutional networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 4700-4708.

[20] Xie J, He N, Fang L, et al. Scale-free convolutional neural network for remote sensing scene classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(9): 6916-6928.

[21] Liu X, Zhou Y, Zhao J, et al. Siamese convolutional neural networks for remote sensing scene classification[J]. IEEE Geoscience and Remote Sensing Letters, 2019, 16(8): 1200-1204.

[22] Fang Z, Li W, Zou J, et al. Using CNN-based high-level features for remote sensing scene classification[C]//2016 IEEE International Geoscience and Remote Sensing Symposium (IGARSS). IEEE, 2016: 2610-2613.

[23] Cheng G, Li Z, Yao X, et al. Remote sensing image scene classification using bag of convolutional features[J]. IEEE Geoscience and Remote Sensing Letters, 2017, 14(10): 1735-1739.

[24] Wang D, Lan J. A deformable convolutional neural network with spatial-channel attention for remote sensing scene classification[J]. Remote Sensing, 2021, 13(24): 5076.

[25] Wang X, Duan L, Shi A, et al. Multilevel feature fusion networks with adaptive channel dimensionality reduction for remote sensing scene classification[J]. IEEE Geoscience and Remote Sensing Letters, 2021, 19: 1-5.

[26] Chu J, Zhao G H. Scene classification based on SIFT combined with GIST[C]//2014 International Conference on Information Science, Electronics and Electrical Engineering. IEEE, 2014, 1: 331-336.

[27] Luo B, Jiang S, Zhang L. Indexing of remote sensing images with different resolutions by multiple features[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2013, 6(4): 1899-1912.

[28] Yuan B, Han L, Gu X, et al. Multi-deep features fusion for high-resolution remote sensing image scene classification[J]. Neural Computing and Applications, 2021, 33: 2047-2063.

[29] Ji J, Zhang T, Jiang L, et al. Combining multilevel features for remote sensing image scene classification with attention model[J]. IEEE Geoscience and Remote Sensing Letters, 2019, 17(9): 1647-1651.

[30] Liu Y, Zhong Y, Qin Q. Scene classification based on multiscale convolutional neural network[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(12): 7109-7121.

[31] 吴晨, 王宏伟, 袁昱纬, 等. 基于图像特征融合的遥感场景零样本分类算法[J]. 光学学报, 2019, 39(6): 98-110.

[32] 张艳月, 张宝华, 赵云飞, 等. 基于双通道深度密集特征融合的遥感影像分类[J]. 激光技术, 2021, 45(1): 73-79.

[33] 张珂, 冯晓晗, 郭玉荣, 等. 图像分类的深度卷积神经网络模型综述[J]. 中国图象图形学报, 2021, 26(10): 2305-2325.

[34] Hu J, Shen L, Sun G. Squeeze-and-excitation networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018: 7132-7141.

[35] Wang Q, Wu B, Zhu P, et al. ECA-Net: Efficient channel attention for deep convolutional neural networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2020: 11534-11542.

[36] Li X, Wang W, Hu X, et al. Selective kernel networks[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2019: 510-519.

[37] Woo S, Park J, Lee J Y, et al. CBAM: Convolutional block attention module[C]//Proceedings of the European Conference on Computer Vision (ECCV). 2018: 3-19.

[38] 王燕, 李国臣, 孙晓丽. 基于多分类器融合的高光谱图像分类研究[J]. 兰州理工大学学报, 2022, 48(1): 98-106.

[39] 甘富文, 武明辉, 吴亚平, 等. 多特征融合的肝细胞癌分化等级术前预测方法研究[J]. 计算机应用与软件, 2022, 39(7): 147-153.

[40] Cheng G, Han J, Lu X. Remote sensing image scene classification: Benchmark and state of the art[J]. Proceedings of the IEEE, 2017, 105(10): 1865-1883.

[41] Xia G S, Hu J, Hu F, et al. AID: A benchmark data set for performance evaluation of aerial scene classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2017, 55(7): 3965-3981.

[42] Zhou W, Newsam S, Li C, et al. PatternNet: A benchmark dataset for performance evaluation of remote sensing image retrieval[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2018, 145: 197-209.

[43] Deng J, Dong W, Socher R, et al. Imagenet: A large-scale hierarchical image database[C]//2009 IEEE Conference on Computer Vision and Pattern Recognition. IEEE, 2009: 248-255.

[44] Yao H, Chuyi L, Dan H, et al. Gabor feature based convolutional neural network for object recognition in natural scene[C]//2016 3rd International Conference on Information Science and Control Engineering (ICISCE). IEEE, 2016: 386-390.

[45] Tao Z, Wei T, Li J. Wavelet multi-level attention capsule network for texture classification[J]. IEEE Signal Processing Letters, 2021, 28: 1215-1219.

[46] 郭锦熙. 基于特征提取与 CNN 模型融合的海洋鱼类识别[D]. 天津: 天津工业大学, 2019.

[47] Yu Y, Liu F. A two-stream deep fusion framework for high-resolution aerial scene classification[J]. Computational Intelligence and Neuroscience, 2018: 1-13.

[48] He N, Fang L, Li S, et al. Skip-connected covariance network for remote sensing scene classification[J]. IEEE Transactions on Neural Networks and Learning Systems, 2019, 31(5): 1461-1474.

[49] Xu K, Huang H, Li Y, et al. Multilayer feature fusion network for scene classification in remote sensing[J]. IEEE Geoscience and Remote Sensing Letters, 2020, 17(11): 1894-1898.

[50] Tian T, Li L, Chen W, et al. SEMSDNet: A Multi-Scale Dense Network with Attention for Remote Sensing Scene Classification[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021, 14: 5501–5514.

[51] Zhang G, Xu W, Zhao W, et al. A multiscale attention network for remote sensing scene images classification[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021, 14: 9530-9545.

[52] Gao Y, Shi J, Li J, et al. Remote sensing scene classification based on high-order graph convolutional network[J]. European Journal of Remote Sensing, 2021, 54(1): 141-155.

[53] Shi C, Zhang X, Wang L. A lightweight convolutional neural network based on channel multi-group fusion for remote sensing scene classification[J]. Remote Sensing, 2022, 14(1): 9.

[54] Yang S, Song F, Jeon G, et al. Scene changes understanding framework based on graph convolutional networks and swin transformer blocks for monitoring LCLU using high-resolution remote sensing images[J]. Remote Sensing, 2022, 14(15): 3709.

[55] Wang X, Wang S, Ning C, et al. Enhanced feature pyramid network with deep semantic embedding for remote sensing scene classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 59(9): 7918-7932.

[56] Shen J, Zhang T, Wang Y, et al. A dual-model architecture with grouping-attention-fusion for remote sensing scene classification[J]. Remote Sensing, 2021, 13(3): 433.

[57] Shen J, Yu T, Yang H, et al. An attention cascade global–local network for remote sensing scene classification[J]. Remote Sensing, 2022, 14(9): 2042.

[58] Deepan P, Sudha L R. Remote sensing image scene classification using dilated convolutional neural networks[J]. International Journal, 2020, 8(7): 3622-3630.

[59] Shafaey M A, Salem M A M, Ebeid H M, et al. Comparison of CNNs for remote sensing scene classification[C]//2018 13th International Conference on Computer Engineering and Systems (ICCES). IEEE, 2018: 27-32.

[60] Guo Y, Ji J, Lu X, et al. Global-local attention network for aerial scene classification[J]. IEEE Access, 2019, 7: 67200-67212.

[61] Guo D, Xia Y, Luo X. Scene classification of remote sensing images based on saliency dual attention residual network[J]. IEEE Access, 2020, 8: 6344-6357.

[62] Tian Q, Wan S, Jin P, et al. A novel feature fusion with self-adaptive weight method based on deep learning for image classification[C]//Advances in Multimedia Information Processing–PCM 2018: 19th Pacific-Rim Conference on Multimedia, Hefei, China, September 21-22, 2018, Proceedings, Part I 19. Springer International Publishing, 2018: 426-436.