遥感图像变化检测是遥感图像解译领域中的研究热点之一，其目的在于快速准确地对同一地点不同时相获取的两幅遥感图像进行分析处理，进而寻找出其中所发生的变化信息，在自然灾害监测、环境监测、城市规划等方面具有非常重要的研究意义和应用价值。因此，本文以合成孔径雷达（Synthetic Aperture Radar，SAR）图像为研究对象，针对图像高维度特征的最优筛选与融合、高阶空间结构信息的全面捕获与挖掘、分类器判别能力和抗噪能力的有效提高等关键问题，以随机场模型为基础理论，对遥感图像变化检测技术展开研究。本论文的研究内容主要概括为以下三个部分：

（1）针对遥感图像中高维度特征带来的特征冗余问题，本文提出了一种双层随机森林网络（Two-Layer Random Forest，TL-RF）模型用于变化检测，在该网络的第一层随机森林结构中，TL-RF根据重要性系数对高维度特征进行排序，筛选出较为重要的特征集合作为第二层结构的输入，并在第二层随机森林结构中进行决策分类，进而实现更有效的特征融合。实验结果表明，相比较传统的随机森林算法，TL-RF所得到的变化检测精度有了很大的提高。

（2）基于混合判别随机场（Hybrid Discriminative Random Fields，HDRF）模型在遥感图像分析领域中的优势，本文将提出的TL-RF模型引入HDRF模型中，利用TL-RF模型输出的类条件概率对一元势能函数进行建模，二元势能函数通过广义伊辛模型进行构建，统计分布则采用具有较高普适性的广义伽马分布进行建模。实验结果表明，TL-RF模型的引入有效地提高了HDRF模型的鲁棒性和抗噪性。

（3）针对SAR图像复杂空间结构信息的有效捕获，本文利用高阶结构对HDRF模型进行优化，提出了基于RF-HoDRF（High-Order Hybrid Discriminative Random Field Improved by Two-Layer Random Forest）模型的变化检测算法，该模型利用鲁棒的Pⁿ模型对高阶势能进行建模，能够同时捕获判别性纹理特征，空间邻域信息，统计分布特性以及高阶结构信息。实验结果表明，RF-HoDRF模型能够更好地保持区域的光滑性和一致性，能够得到更高的变化检测精度。

Remote sensing image change detection is one of the research hotspots in the field of remote sensing image interpretation. Its purpose is to quickly and accurately analyse and process two remote sensing images acquired at different time phases of the same location, and then find out the change information occurring in them, which has very important research significance and application value in natural disaster monitoring, environmental monitoring, urban planning and so on. Therefore, this thesis takes Synthetic Aperture Radar (SAR) images as the research object, and addresses the key issues of optimal screening and fusion of high-dimensional features, comprehensive capture and mining of higher-order spatial structure information, and effective improvement of classifier discriminatory ability and noise immunity, etc., and uses the random field model as the basic theory to investigate the change detection technology of remote sensing images. The research content of this thesis is mainly summarized as the following three parts:

(1) To address the problem of feature redundancy caused by high-dimensional features in remote sensing images, this thesis proposes a Two-Layer Random Forest (TL-RF) model for change detection. In the second layer of the random forest structure, TL-RF ranks the high-dimensional features according to their importance coefficients and selects the more important sets of features as input to the second layer of the structure, and performs decision classification in the second layer of the random forest structure to achieve more effective feature fusion. The experimental results show that the change detection accuracy obtained by TL-RF has been greatly improved compared to the traditional random forest algorithm.

(2) Based on the advantages of the Hybrid Discriminative Random Fields (HDRF) model in the field of remote sensing image analysis, the proposed TL-RF model is introduced into the HDRF model in this thesis, and the monadic potential function is modelled using the class of conditional probabilities output from the TL-RF model, the binary potential function is modelled through the generalized Ising model is constructed, and the statistical distribution is modelled using the generalized gamma distribution with high generality. The experimental results show that the introduction of the TL-RF model effectively improves the robustness and noise immunity of the HDRF model.

(3) To effectively capture the complex spatial structure information of SAR images, this thesis optimizes the HDRF model by using high-order structures and proposes a change detection algorithm based on the RF-HoDRF (High-Order Hybrid Discriminative Random Field Improved by Two-Layer Random Forest) model. The change detection algorithm is based on the RF-HoDRF model, which uses a robust Pⁿ model to model the higher-order potentials and is able to capture discriminative texture features, spatial neighborhoods information, statistical distribution properties and higher-order structure information simultaneously. The experimental results show that the RF-HoDRF model can better maintain the smoothness and consistency of the region, and can obtain higher change detection accuracy.

[1]朱述龙, 张占睦, 遥感图像获取与分析[M]. 北京: 科学出版社, 2000.

[2]保铮, 邢孟道, 王彤编著. 雷达成像技术[M]. 北京: 电子工业出版社, 2005.

[3]Zhu X X, Tuia D, Mou L, et al. Deep learning in remote sensing: A comprehensive review and list of resources[J]. IEEE Geoscience and Remote Sensing Magazine, 2017, 5(4): 8-36.

[4]Shi W, Zhang M, Zhang R, et al. Change detection based on artificial intelligence: State-of-the-art and challenges[J]. Remote Sensing, 2020, 12(10): 1688.

[5]Tsokas A, Rysz M, Pardalos P M, et al. SAR data applications in earth observation: An overview[J]. Expert Systems with Applications, 2022: 117342.

[6]Saha S, Bovolo F, Bruzzone L. Building change detection in VHR SAR images via unsupervised deep transcoding[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 59(3): 1917-1929.

[7]Li N , Niu S , Guo Z , et al. Dynamic Waterline Mapping of Inland Great Lakes Using Time-Series SAR Data From GF-3 and S-1A Satellites: A Case Study of DJK Reservoir, China[J]. IEEE journal of selected topics in applied earth observations and remote sensing, 2019(12-11).

[8]Stephenson O L, Köhne T, Zhan E, et al. Deep learning-based damage mapping with InSAR coherence time series[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 60: 1-17.

[9]Camps-Valls G, Gomez-Chova L, Munoz-Mari J, et al. Kernel-Based Framework for Multitemporal and Multisource Remote Sensing Data Classification and Change Detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2008, 46(6):1822-1835.

[10]Bovolo F, Bruzzone L, Marconcini M. A Novel Approach to Unsupervised Change Detection Based on a Semisupervised SVM and a Similarity Measure[J]. IEEE Transactions on Geoscience and Remote Sensing, 2008, 46(7):2070-2082.

[11]Jia L, Li M, Wu Y, et al. Semisupervised SAR image change detection using a cluster-neighborhood kernel[J]. IEEE Geoscience and Remote Sensing Letters, 2014, 11(8): 1443-1447.

[12]Jia L, Li M, Zhang P, et al. SAR image change detection based on correlation kernel and multistage extreme learning machine[J]. IEEE Transactions on Geoscience and Remote sensing, 2016, 54(10): 5993-6006.

[13]Tan X, Li M, Zhang P, et al. Multispectral image change detection with kernel cross-modal factor analysis-based fusion of kernels[J]. Journal of Applied Remote Sensing, 2018, 12(3): 035008-035008.

[14]Jia L, Zhang T, Fang J, et al. Multiple kernel graph cut for SAR image change detection[J]. Remote Sensing, 2021, 13(4): 725.

[15]Peng D, Zhang Y, Guan H. End-to-end change detection for high resolution satellite images using improved UNet plus[J]. Remote Sensing, 2019, 11(11): 2072-4292

[16]Fu Q, Luo W L, Lü J X. Land utilization change detection of satellite remote sensing image based on AlexNet and support vector machine[J]. Laser and Optoelectronics Progress, 2020, 57(17): 172802.

[17]Zhang P, Gong M, Su L, et al. Change detection based on deep feature representation and mapping transformation for multi-spatial-resolution remote sensing images[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2016, 116: 24-41.

[18]Gong M, Zhao J, Liu J, et al. Change Detection in Synthetic Aperture Radar Images Based on Deep Neural Networks[J]. IEEE Transactions on Neural Networks & Learning Systems, 2017, 27(1):125-138.

[19]Luo B, Hu C, Su X, et al. Differentially deep subspace representation for unsupervised change detection of SAR images[J]. Remote Sensing, 2019, 11(23): 2740.

[20]Gao Y, Gao F, Dong J, et al. Change Detection From Synthetic Aperture Radar Images Based on Channel Weighting-Based Deep Cascade Network[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2019, PP(99):1-13.

[21]Saha S, Bovolo F, Bruzzone L. Building Change Detection in VHR SAR Images via Unsupervised Deep Transcoding[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, PP(99):1-13.

[22]Zhang X, Su H, Zhang C, et al. Robust unsupervised small area change detection from SAR imagery using deep learning[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2021, 173: 79-94.

[23]Wang R, Wang L, Wei X, et al. Dynamic graph-level neural network for SAR image change detection[J]. IEEE Geoscience and Remote Sensing Letters, 2021, 19: 1-5.

[24]周志华著, 李楠译. 集成学习[M]. 北京: 电子工业出版社, 2020.

[25]周志华著. 机器学习[M]. 北京: 清华大学出版社, 2016.

[26]Li H, Ma J, Gong M, et al. Change detection in synthetic aperture radar images based on evolutionary multiobjective optimization with ensemble learning[J]. Memetic Computing, 2015, 7: 275-289.

[27]Cui B, Zhang Y, Yan L, et al. An unsupervised SAR change detection method based on stochastic subspace ensemble learning[J]. Remote Sensing, 2019, 11(11): 1314.

[28]Zerrouki N, Harrou F, Sun Y, et al. A Machine Learning-Based Approach for Land Cover Change Detection Using Remote Sensing and Radiometric Measurements[J]. IEEE Sensors Journal, 2019, PP(14):1-1.

[29]刘国英, 马国锐, 王雷光等著. 基于Markov随机场的小波域图像建模及分割Matlab环境[M]. 北京: 科学出版社, 2010.

[30]Gong M, Su L, Jia M, et al. Fuzzy clustering with a modified MRF energy function for change detection in synthetic aperture radar images[J]. IEEE Transactions on Fuzzy Systems, 2013, 22(1): 98-109.

[31]Solarna D, Moser G, Serpico S B. A Markovian Approach to Unsupervised Change Detection with Multiresolution and Multimodality SAR Data[J]. Remote Sensing, 2018, 10(11):1671-.

[32]Fang H, Du P, Wang X, et al. Unsupervised change detection based on weighted change vector analysis and improved Markov random field for high spatial resolution imagery[J]. IEEE Geoscience and Remote Sensing Letters, 2021, 19: 1-5.

[33]Lafferty J, Mccallum A, Pereira F C N. Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data[J]. proceedings of ICML, 2001.

[34]Li H, Li M, Zhang P, et al. SAR image change detection based on hybrid conditional random field[J]. IEEE Geoscience and Remote Sensing Letters, 2014, 12(4): 910-914.

[35]Zhou L, Cao G, Li Y, et al. Change detection based on conditional random field with region connection constraints in high-resolution remote sensing images[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2016, 9(8): 3478-3488.

[36]Sun Y, Lei L, Guan D, et al. Iterative structure transformation and conditional random field based method for unsupervised multimodal change detection[J]. Pattern Recognition, 2022, 131: 108845.

[37]Kumar S, Hebert M. Discriminative random fields[J]. International Journal of Computer Vision, 2006, 68: 179-201.

[38]An L, Li M, Zhang P, et al. Discriminative random fields based on maximum entropy principle for semisupervised SAR image change detection[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2015, 9(8): 3395-3404.

[39]Song W, Wu Y, Guo P. Composite Kernel and Hybrid Discriminative Random Field Model Based on Feature Fusion for PolSAR Image Classification[J]. IEEE Geoscience and Remote Sensing Letters, 2020, PP(99):1-5.

[40]Zhang P, Li M, Wu Y, et al. Unsupervised SAR image segmentation using high-order conditional random fields model based on product-of-experts[J]. Pattern Recognition Letters, 2016, 78: 48-55.

[41]Sulimowicz L, Ahmad I, Aved A. Superpixel-enhanced pairwise conditional random field for semantic segmentation[C]//2018 25th IEEE International Conference on Image Processing (ICIP). IEEE, 2018: 271-275.

[42]Dong H, Ma W, Jiao L, et al. A multiscale self-attention deep clustering for change detection in SAR images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 60: 1-16.

[43]宋婉莹. 基于统计分布和随机场模型的极化SAR图像分类算法研究[D]. 西安: 西安电子科技大学, 2018.

[44]Daugman J G. Uncertainty relation for resolution in space, spatial frequency, and orientation optimized by two-dimensional visual cortical filters[J]. Journal of the Optical Society of America A, 1985, 2(7): 1160-1169.

[45]Breiman L. Random forests[J]. Machine Learning, 2001, 45: 5-32.

[46]Sales M H R, De Bruin S, Souza C, et al. Land use and land cover area estimates from class membership probability of a random forest classification[J]. IEEE Transactions on Geoscience and Remote Sensing, 2021, 60: 1-11.

[47]Sheykhmousa M, Mahdianpari M, Ghanbari H, et al. Support vector machine versus random forest for remote sensing image classification: A meta-analysis and systematic review[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2020, 13: 6308-6325.

[48]Zhang P, Li M, Wu Y, et al. Hierarchical conditional random fields model for semisupervised SAR image segmentation[J]. IEEE Transactions on Geoscience and Remote Sensing, 2015, 53(9): 4933-4951.

[49]Hoberg T, Rottensteiner F, Feitosa R Q, et al. Conditional random fields for multitemporal and multiscale classification of optical satellite imagery[J]. IEEE Transactions on Geoscience and Remote Sensing, 2014, 53(2): 659-673.

[50]Lv P, Zhong Y, Zhao J, et al. Change detection based on a multifeature probabilistic ensemble conditional random field model for high spatial resolution remote sensing imagery[J]. IEEE Geoscience and Remote Sensing Letters, 2016, 13(12): 1965-1969.

[51]Goodman J W. Some fundamental properties of speckle[J]. Journal of the Optical Society of America, 1976, 66(11): 1145-1150.

[52]李永晨, 刘浏. SAR 图像统计模型综述[J]. 计算机工程与应用, 2013, 49(13): 180-186.

[53]Song W, Wu Y, Guo P. Composite kernel and hybrid discriminative random field model based on feature fusion for PolSAR image classification[J]. IEEE Geoscience and Remote Sensing Letters, 2020, 18(6): 1069-1073.

[54]Zhang P, Boudaren M E Y, Jiang Y, et al. High-order triplet CRF-PCANet for unsupervised segmentation of nonstationary SAR image[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 59(10): 8433-8454.

[55]Song W, Wu Y, Xiao X. Nonstationary PolSAR image classification by deep-features-based high-order triple discriminative random field[J]. IEEE Geoscience and Remote Sensing Letters, 2020, 18(8): 1406-1410.

[56]Achanta R, Shaji A, Smith K, et al. SLIC superpixels compared to state-of-the-art superpixel methods[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(11): 2274-2282.

[57]王灵矫, 钟益群, 郭华, 彭志强. 基于高阶条件随机场模型的改进型图像分割算法[J]. 计算机工程,2016,42(06):241-246.