-
摘要:
为提高卷积神经网络(CNN)在变化检测任务中的鲁棒性能和泛化能力，提出一种基于拉普拉斯支持向量机(LapSVM)的卷积小波神经网络(CWNLSN)模型。通过样本标记方法，获得高置信度的“伪标签”，划分出网络训练集、分类训练集和待测集；在CNN中，引入离散小波池化，用于提取局部空间深度特征；设计基于LapSVM的局部空间深度特征分类(LSDC)模块，对特征进行分类，判别待测集的变化情况；设置对比实验和消融实验，在多组真实遥感数据集上进行测试。结果表明，所提方法获得了更显著的变化检测效果。
Abstract:This research offers an image change detection strategy based on convolutional-wavelet neural networks based on Laplace support vector machine (LapSVM) (CWNLSN) to enhance the generalization and robustness of convolutional neural network (CNN) in change detection applications. Firstly, by using the sample labeling method, high confidence "pseudo labels" are obtained, and the network training set, classification training set, and test set are divided. Secondly, discrete wavelet pooling is used to retrieve local space deep characteristics in CNN. Then, design a local space deep feature classification (LSDC) module based on LapSVM to classify the deep features and distinguish the changed information in the test set. Finally, comparative experiments and ablation experiments were conducted on multiple sets of real remote sensing datasets for testing. The results indicate that the proposed method achieved a more significant change detection effect.
-
图 3 基于LapSVM的LSDC模块的投票方法执行流程
Figure 3. The execution process of voting method for local space deep feature classification module based on LapSVM
图 4 基于LapSVM的LSDC模块的均值方法执行流程
Figure 4. The execution process of the mean method for local space deep feature classification module based on LapSVM
图 5 CWNLSN-vote投票结果准确率随邻域大小的变化曲线
Figure 5. Curves of the accuracy of CWNLSN-vote with neighborhood size
图 6 在旧金山数据集、黄河海岸线数据集、黄河农田数据集、贵州数据集中不同变化检测方法的可视化结果
Figure 6. Visualization results of different change detection methods in the San Francisco dataset, Yellow River coastline dataset, Yellow River farmland dataset, and Guizhou dataset
图 7 旧金山数据集上不同变化检测方法的Kappa系数和F1分数随邻域大小的变化
Figure 7. Kappa coefficients and F1 scores for different change detection methods on the San Francisco dataset as a function of neighborhood size
图 8 黄河农田数据集上不同变化检测方法的Kappa系数和F1分数随邻域大小的变化
Figure 8. Kappa coefficients and F1 scores for different change detection methods on the Yellow River farmland dataset as a function of neighborhood size
图 9 贵州数据集上不同变化检测方法的Kappa系数和F1分数随邻域大小的变化
Figure 9. Kappa coefficients and F1 scores of different change detection methods on the Guizhou dataset as a function of neighborhood size
图 10 不同数据集上不同变化检测方法的平均Kappa系数和F1分数
Figure 10. Average Kappa coefficients and F1 scores for different change detection methods on different datasets
表 1 CNN及其改进模型
Table 1. CNN model and improved CNN models
序号 模型 池化方式 特征分类方式 1 CNN 最大池化 激活函数 2 CDWNN 双数复小波池化 激活函数 3 CWNN 离散小波池化 激活函数 4 CNLSN 最大池化 LapSVM 5 CDWNLSN 双数复小波池化 LapSVM 6 CWNLSN 离散小波池化 LapSVM 
下载: 导出CSV
表 2 对比实验中多个数据集上变化检测结果的客观评估
Table 2. Objective evaluation of change detection results on multiple datasets in comparative experiments
方法 错检数 漏检数 准确率/% 旧金山 黄河海岸线 黄河农田 贵州 旧金山 黄河海岸线 黄河农田 贵州 旧金山 黄河海岸线 黄河农田 贵州 PCAK[25] 2 047 0 490 99 4033 38217 1279 38768 94.47 69.67 98.01 75.71 PCANet[24] 5122 4 1345 106 720 28335 201 31990 94.69 77.51 98.26 79.94 CDWNN[14] 226 51 734 383 13647 20258 969 19069 87.39 83.88 98.09 87.84 TFCDWNN[18] 5849 137 811 324 54 152 213 701 94.63 99.77 98.85 99.36 SCapsNet[12] 4241 171 771 450 723 96 173 451 95.49 99.79 98.94 99.44 CWNLSN-vote 3929 192 473 582 592 61 560 153 95.95 99.80 98.84 99.54 方法 F1分数/% Kappa系数/% 用时/s 旧金山 黄河海岸线 黄河农田 贵州 旧金山 黄河海岸线 黄河农田 贵州 旧金山 黄河海岸线 黄河农田 贵州 PCAK[25] 79.68 6.59 84.39 75.71 76.50 4.62 83.33 4.98 0.3 0.5 0.2 0.7 PCANet[24] 75.18 8.66 83.54 79.94 72.32 6.76 82.64 6.28 1262.3 2491.7 994.2 2717.5 CDWNN[14] 66.46 11.33 84.19 87.84 59.67 9.50 83.18 8.65 607 618 590.8 642.1 TFCDWNN[18] 73.34 89.34 89.70 99.36 70.58 89.22 89.09 69.18 473.5 492.4 446.3 225 SCapsNet[12] 79.67 89.81 90.50 99.44 77.19 89.71 89.95 69.53 99.8 84.1 204.4 248.6 CWNLSN-vote 81.83 90.14 90.28 99.54 79.60 90.04 89.66 71.01 603.4 598.2 579.2 692.5
下载: 导出CSV
表 3 消融实验中多个数据集上变化检测结果的客观评估
Table 3. Objective evaluation of change detection results on multiple datasets in ablation experiments
方法 错检数 漏检数 准确率/% F1分数/% Kappa系数/% 旧金山 黄河农田 贵州 旧金山 黄河农田 贵州 旧金山 黄河农田 贵州 旧金山 黄河农田 贵州 旧金山 黄河农田 贵州 CNN 4378 864 463 515 239 603 95.55 98.76 99.33 79.68 88.88 65.88 77.25 88.22 65.54 CWNN 4284 852 416 500 215 515 95.65 98.80 99.42 80.19 89.23 69.80 77.82 88.59 69.51 CNLSN-mean 3996 498 622 559 575 141 95.86 98.79 99.52 81.39 89.89 69.46 79.11 89.25 69.23 CWNLSN-mean 3837 489 579 559 534 154 96.00 98.85 99.54 82.15 90.34 71.35 79.95 89.73 71.13
下载: 导出CSV
-
[1] 杨刚. 基于特征学习与聚类分析的SAR图像变化检测研究[D]. 成都: 西南交通大学, 2021.YANG G. SAR image change detection based on feature learning and clustering analysis[D]. Chengdu: Southwest Jiaotong University, 2021(in Chinese). [2] 刘国强, 房胜, 李哲. 用于遥感图像变化检测的全尺度特征聚合网络[J]. 北京航空航天大学学报, 2022, 48(8): 1464-1470.LIU G Q, FANG S, LI Z. A full-scale feature aggregation network for remote sensing image change detection[J]. Journal of Beijing University of Aeronautics and Astronautics, 2022, 48(8): 1464-1470(in Chinese). [3] 周圆, 李祥瑞, 杨晶. 基于混合网络的异源遥感图像变化检测[J]. 北京航空航天大学学报, 2021, 47(3): 451-460.ZHOU Y, LI X R, YANG J. Heterogeneous remote sensing image change detection based on hybrid network[J]. Journal of Beijing University of Aeronautics and Astronautics, 2021, 47(3): 451-460(in Chinese). [4] SUN Y L, LEI L, LI X, et al. Nonlocal patch similarity based heterogeneous remote sensing change detection[J]. Pattern Recognition, 2021, 109: 107598. [5] NAVARRO A, ROLIM J, MIGUEL I, et al. Crop monitoring based on SPOT-5 Take-5 and Sentinel-1A data for the estimation of crop water requirements[J]. Remote Sensing, 2016, 8(6): 525. [6] OTHMAN A A, GLOAGUEN R. River courses affected by landslides and implications for hazard assessment: a high resolution remote sensing case study in NE Iraq-W Iran[J]. Remote Sensing, 2013, 5(3): 1024-1044. [7] GONG M G, ZHOU Z Q, MA J J. Change detection in synthetic aperture radar images based on image fusion and fuzzy clustering[J]. IEEE Transactions on Image Processing, 2012, 21(4): 2141-2151. [8] BEZDEK J C. Pattern recognition with fuzzy objective function algorithms[M]. New York: Plenum Press, 1981. [9] GONG M G, ZHAO J J, LIU J, et al. Change detection in synthetic aperture radar images based on deep neural networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2016, 27(1): 125-138. [10] LI M K, LI M, ZHANG P, et al. SAR image change detection using PCANet guided by saliency detection[J]. IEEE Geoscience and Remote Sensing Letters, 2019, 16(3): 402-406. [11] WANG S N, WANG Y N, LIU Y, et al. SAR image change detection based on sparse representation and a capsule network[J]. Remote Sensing Letters, 2021, 12(9): 890-899. [12] 张益天, 罗喜伶, 王宇鹏. 基于轻量胶囊网络的自监督图像变化检测方法[J]. 北京航空航天大学学报, 2025, 51(5): 1705-1715.ZHANG Y T, LUO X L, WANG Y P. Self-supervised image change detection method based on lightweight capsule network[J]. Journal of Beijing University of Aeronautics and Astronautics, 2025, 51(5): 1705-1715(in Chinese). [13] LIU F, JIAO L C, TANG X, et al. Local restricted convolutional neural network for change detection in polarimetric SAR images[J]. IEEE Transactions on Neural Networks and Learning Systems, 2019, 30(3): 818-833. [14] GAO F, WANG X, GAO Y H, et al. Sea ice change detection in SAR images based on convolutional-wavelet neural networks[J]. IEEE Geoscience and Remote Sensing Letters, 2019, 16(8): 1240-1244. [15] WANG Q, YUAN Z H, DU Q, et al. GETNET: a general end-to-end 2-D CNN framework for hyperspectral image change detection[J]. IEEE Transactions on Geoscience and Remote Sensing, 2019, 57(1): 3-13. [16] 刘过. 基于深度学习的SAR图像变化检测研究[D]. 重庆: 重庆大学, 2020.LIU G. Study on deep learning-based SAR images change detection[D]. Chongqing: Chongqing University, 2020(in Chinese). [17] D’HAEYER J P F. Gaussian filtering of images: a regularization approach[J]. Signal Processing, 1989, 18(2): 169-181. [18] ZHANG X Z, 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. [19] BOVOLO F, BRUZZONE L. A detail-preserving scale-driven approach to change detection in multitemporal SAR images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2005, 43(12): 2963-2972. [20] OTSU N. A threshold selection method from gray-level histograms[J]. IEEE Transactions on Systems, Man, and Cybernetics, 1979, 9(1): 62-66. [21] MALLAT S G. A theory for multiresolution signal decomposition: the wavelet representation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1989, 11(7): 674-693. [22] DUAN Y P, LIU F, JIAO L C, et al. SAR image segmentation based on convolutional-wavelet neural network and Markov random field[J]. Pattern Recognition, 2017, 64: 255-267. [23] CHAN T H, JIA K, GAO S H, et al. PCANet: a simple deep learning baseline for image classification[J]. IEEE Transactions on Image Processing, 2015, 24(12): 5017-5032. [24] GAO F, DONG J Y, LI B, et al. Automatic change detection in synthetic aperture radar images based on PCANet[J]. IEEE Geoscience and Remote Sensing Letters, 2016, 13(12): 1792-1796. [25] CELIK T. Unsupervised change detection in satellite images using principal component analysis and K-means clustering[J]. IEEE Geoscience and Remote Sensing Letters, 2009, 6(4): 772-776. [26] BAZI Y, BRUZZONE L, MELGANI F. An unsupervised approach based on the generalized Gaussian model to automatic change detection in multitemporal SAR images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2005, 43(4): 874-887. [27] ZHANG X Z, LIU G, ZHANG C, et al. Two-phase object-based deep learning for multi-temporal SAR image change detection[J]. Remote Sensing, 2020, 12(3): 548. -
下载:
点击查看大图
计量
- 文章访问数: 306
- HTML全文浏览量: 84
- PDF下载量: 16
- 被引次数: 0
