Remote Sensing for Natural Resources >
Multivariate alteration detection using angle thresholds based on a context-sensitive Bayesian network
Received date: 2024-08-07
Revised date: 2024-11-12
Online published: 2026-06-03
In the field of multivariate alteration detection (MAD) of remote sensing images,change vector analysis in posterior probability space (CVAPS) is a widely used method. However,the CVAPS,which employs support vector machines to estimate the posterior probability vectors of remote sensing image pixels,is susceptible to various factors such as different objects with the same spectrum,the same object with different spectra,and mixed pixels in remote sensing images. These factors make it difficult to accurately estimate the magnitude and direction of the posterior probability vectors of complex pixels,consequently affecting the accuracy of multivariate alteration detection. Therefore,under the framework of CVAPS,this paper proposed a MAD method using angle thresholds,which employed the fuzzy C-means clustering to decompose mixed pixels and coupled a context-sensitive Bayesian network. When the angle is less than a certain threshold,the pixel is identified as the change type represented by the standard change vector. Experimental results show that the proposed algorithm exhibited superior alteration detection performance,achieving higher change detection accuracy than other algorithms.
ZHU Rui , LI Yikun , LI Xiaojun , YANG Shuwen , XIE Jiangling . Multivariate alteration detection using angle thresholds based on a context-sensitive Bayesian network[J]. Remote Sensing for Natural Resources, 2025 , 37(5) : 131 -140 . DOI: 10.6046/zrzyyg.2024267
P((m,n)|Lv)≈
P((m,n)|Lv,Tv)=表1 各研究区地物类型数量Tab.1 Number of feature types in each study area (个) |
| 地物(变化)类型 | 研究区1 | 研究区2 | 研究区3 |
|---|---|---|---|
| 总地物类型 | 3 | 4 | 5 |
| 可能变化类型 | 6 | 12 | 20 |
| 实际变化类型 | 1 | 2 | 3 |
表2 研究区1和研究区2变化检测算法性能比较Tab.2 Performance results of change detection algorithms in study area 1 and study area 2 |
| 算法 | 研究区1(林地到荒地) | 研究区2(建筑物到林地) | 研究区2(建筑物到荒地) | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 错检 率/% | 漏检 率/% | 总体精 度/% | Kappa 系数 | 错检 率/% | 漏检 率/% | 总体精 度/% | Kappa 系数 | 错检 率/% | 漏检 率/% | 总体精 度/% | Kappa 系数 | |
| 本文算法 | 13.08 | 2.62 | 94.69 | 0.879 3 | 21.26 | 21.38 | 98.46 | 0.778 8 | 22.34 | 13.44 | 98.09 | 0.808 7 |
| FCM-SBN- CVAPS-AT | 26.61 | 4.37 | 87.98 | 0.739 9 | 58.69 | 7.77 | 95.00 | 0.548 1 | 18.16 | 33.18 | 97.81 | 0.723 4 |
| SVM-CVAPS- AT | 18.61 | 14.08 | 89.62 | 0.760 1 | 50.24 | 54.42 | 96.38 | 0.497 1 | 49.49 | 44.89 | 95.08 | 0.501 2 |
| FCM-CSBN- CVAPS-MC | 14.82 | 3.36 | 95.62 | 0.859 0 | 26.64 | 28.14 | 98.05 | 0.715 9 | 19.54 | 33.62 | 97.53 | 0.714 6 |
| FCM-SBN- CVAPS-MC | 18.26 | 4.96 | 91.94 | 0.819 0 | 57.97 | 15.04 | 95.75 | 0.510 2 | 18.69 | 43.47 | 97.19 | 0.652 8 |
| SVM-CVAPS- MC | 8.45 | 68.23 | 78.10 | 0.372 1 | —① | — | — | — | 53.21 | 73.84 | 94.85 | 0.311 0 |
| FCM-CSBN- PCC | 24.72 | 1.99 | 89.48 | 0.772 3 | 51.90 | 34.73 | 96.21 | 0.534 6 | 58.04 | 42.18 | 93.93 | 0.454 9 |
| DCVA | 30.86 | 2.78 | 85.84 | 0.701 7 | 90.05 | 9.33 | 70.10 | 0.122 4 | 94.32 | 69.83 | 71.61 | 0.013 0 |
①本节实验中“—”代表精度过低,无参考意义。 |
表3 研究区3变化检测算法性能比较Tab.3 Performance results of change detection algorithms in study area 3 |
| 算法 | 荒地到建筑物 | 建筑物到荒地 | 林地到建筑物 | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 错检 率/% | 漏检 率/% | 总体精 度/% | Kappa 系数 | 错检 率/% | 漏检 率/% | 总体精 度/% | Kappa 系数 | 错检 率/% | 漏检 率/% | 总体精 度/% | Kappa 系数 | |
| 本文算法 | 31.38 | 2.05 | 93.30 | 0.768 1 | 27.96 | 4.76 | 99.45 | 0.817 6 | 26.83 | 18.05 | 99.77 | 0.772 0 |
| FCM-SBN- CVAPS-AT | 28.37 | 22.19 | 92.42 | 0.701 5 | —① | — | — | — | — | — | — | — |
| SVM-CVAPS- AT | 33.95 | 12.20 | 91.80 | 0.705 9 | 48.22 | 16.68 | 98.76 | 0.632 8 | — | — | — | — |
| FCM-CSBN- CVAPS-MC | 8.56 | 63.87 | 91.93 | 0.482 8 | — | — | — | — | — | — | — | — |
| FCM-SBN- CVAPS-MC | 30.39 | 35.53 | 91.75 | 0.622 3 | — | — | — | — | — | — | — | — |
| SVM-CVAPS- MC | 17.81 | 36.56 | 93.96 | 0.682 9 | — | — | — | — | — | — | — | — |
| FCM-CSBN- PCC | 42.86 | 53.06 | 87.38 | 0.443 7 | 93.00 | 86.39 | 96.49 | 0.076 2 | 99.38 | 84.60 | 87.91 | 0.002 9 |
| DCVA | 65.70 | 34.33 | 77.10 | 0.323 5 | 98.00 | 65.13 | 76.63 | 0.013 2 | — | — | — | — |
①本节实验中“—”代表精度过低,无参考意义。 |
| [1] |
李德仁. 利用遥感影像进行变化检测[J]. 武汉大学学报(信息科学版), 2003, 28(s1):7-12.
|
| [2] |
李天宏, 韩鹏. 厦门市土地利用/覆盖动态变化的遥感检测与分析[J]. 地理科学, 2001, 21(6):537-543.
|
| [3] |
严宇, 刘耀林. 基于融合和IFLICM算法的非监督遥感影像变化检测[J]. 测绘通报, 2018(3):25-31.
|
| [4] |
宋嘉鑫, 李轶鲲, 杨树文, 等. 基于后验概率空间变化向量分析的NSCT高分辨率遥感影像变化检测[J]. 自然资源遥感, 2024, 36(3):128-136.doi:10.6046/zrzyyg.2023079.
|
| [5] |
谢江陵, 李轶鲲, 李小军, 等. 基于耦合空间模糊C均值聚类和推土机距离的变化检测[J]. 遥感信息, 2024, 39(3):144-152.
|
| [6] |
|
| [7] |
|
| [8] |
|
| [9] |
|
| [10] |
盛光伟. 基于角度优先变化向量分析的林地变化检测[D]. 南京: 南京大学, 2020.
|
| [11] |
|
| [12] |
|
| [13] |
|
| [14] |
|
| [15] |
龙亦凡, 乔雯钰, 孙静. 基于SVM的大屯矿区遥感影像变化检测[J]. 测绘与空间地理信息, 2020, 43(12):107-110,115.
|
| [16] |
|
| [17] |
李轶鲲, 杨洋, 杨树文, 等. 耦合模糊C均值聚类和贝叶斯网络的遥感影像后验概率空间变化向量分析[J]. 自然资源遥感, 2021, 33(4):82-88.doi:10.6046/zrzyyg.2021032.
|
| [18] |
|
| [19] |
|
| [20] |
|
/
| 〈 |
|
〉 |