基于引力自组织神经网络的震害遥感影像分类

doi:10.6046/gtzyyg.2019.03.13

摘要
图/表
参考文献
相关文章
Metrics

全文: PDF(4443 KB)

HTML
输出: BibTeX | EndNote (RIS)

摘要

针对高空间分辨率遥感影像震害目标难以识别和提取的问题,结合分形纹理和引力自组织神经网络(gravitational self-organizing map,gSOM),提出了一种新的面向对象分类方法。首先,利用分割算法对原始影像进行初始过分割,得到均质性较好的分割单元,以分割单元作为待处理对象; 在此基础上,利用分形纹理描述待分割对象,同时融合光谱特征构建震害目标的特征向量; 最后,利用gSOM对分割对象进行聚类,得到聚类结果,并利用一致性函数以最小代价将多样性的聚类结果集成,最终实现快速、自动决策分类。以四川省汶川县震后高空间分辨率遥感影像为实验数据对算法进行定性和定量的评价,结果表明,该算法能够有效地描述复杂的震害目标,既可以保持大面积震害目标的完整性,也可以反映小的震害目标及其细节信息,提高震害影像的自动分类精度。

	服务

	把本文推荐给朋友
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	黄惠
	郑雄伟
	孙根云
	郝艳玲
	张爱竹
	容俊
	马红章

关键词 ：分形纹理, 引力自组织神经网络, 高空间分辨率遥感, 震害, 面向对象分类

Abstract：

The recognition and extraction of seismic targets in high resolution post-earthquake images pose great challenge to the traditional image classification method. This paper introduces an object-oriented classification method for high resolution post-earthquake images classification, which integrates fractal texture features into a gravitational self-organizing map (gSOM). The method can be summarized as follows. First of all, the mean shift (MS) segmentation algorithm is adopted for initial segmentation in order to obtain homogeneous geographical objects, and the objects are regarded as the basic classification units in the subsequent process. Secondly, the characteristics of objects are quantified by the adaptive combination of the spectral bands and the fractal second order statistics as the texture information extracted from the original seismic image. Finally, the objects as classification units are clustered under the gSOM. For the purpose of controlling the uncertainty in the classification results, these various clustered results are assembled by the consensus function with the least cost. The qualitative and quantitative experiments on the Wenchuan County seismic images demonstrate the effectiveness and accuracy of the proposed method, which not only maintains the integrity of large damage targets, but also reflects details of the small targets at the same time. Also, the method shows the potential in the new technology for high resolution post-event image classification.

Key words： fractal texture gSOM high resolution remote sensing seismic targets object-oriented classification

收稿日期: 2018-06-29 出版日期: 2019-08-30

TP753

基金资助:国家自然科学基金项目“复杂地震环境下多源遥感影像引力智能优化分类模型与算法研究”(41471353);“高异质性滨海湿地盐沼植被环境响应机理与优化分类方法研究”(41801275);山东省自然科学基金项目“引力智能算法引导的高分辨率海岸带城市遥感影像最优分割方法研究”(ZR2018BD007);“光学与微波遥感协同反演植被覆盖区土壤水分研究”(ZR2017MD007);中央高校基本科研业务费专项资金项目“基于多源遥感大数据的“海上丝路”城市不透水面专题信息分层优化提取方法与生态环境效应研究”(18CX05030A);“滨海湿地典型植被群落的高分辨率遥感智能识别研究”(18CX02179A);青岛市博士后应用研究项目“基于超像素的高分辨率海岸带城市影像智能分类方法研究”共同资助(BY20170204)

通讯作者: 孙根云

作者简介: 黄惠(1995-),女,硕士研究生,主要从事遥感影像处理方面的研究。Email: huihuang_rs@163.com.。

引用本文:

黄惠, 郑雄伟, 孙根云, 郝艳玲, 张爱竹, 容俊, 马红章. 基于引力自组织神经网络的震害遥感影像分类[J]. 国土资源遥感, 2019, 31(3): 95-103.
Hui HUANG, Xiongwei ZHENG, Genyun SUN, Yanling HAO, Aizhu ZHANG, Jun RONG, Hongzhang MA. Seismic image classification based on gravitational self-organizing map. Remote Sensing for Land & Resources, 2019, 31(3): 95-103.

链接本文:

https://www.gtzyyg.com/CN/10.6046/gtzyyg.2019.03.13 或 https://www.gtzyyg.com/CN/Y2019/V31/I3/95

特征名称	计算公式	说明
均值	$μ FD = 1 MN ∑ i = 1 M ∑ j = 1 N I FD (i, j)$	计算区域的分形维数的均值
方差	$σ FD = 1 MN ∑ i = 1 M ∑ j = 1 N (I FD (i, j) - μ FD) 2$	计算区域的分形维数的方差
对比度(CON)	$CO N FD = ∑ n = 1 L n 2 [∑ i = 1 M ∑ j = 1 N p ~ (i, j)], \| i - j \| = n$	度量局部区域变化,值越小纹理越均匀
相异性(DIS)	$DI S FD = ∑ i = 1 M ∑ j = 1 N p ~ (i, j) i - j$	描述区域内像素与邻近像素的相异性
空隙度(L)	$L FD = 1 MN ∑ i = 0 M ∑ j = 0 N I FD (i, j) 2 (1 MN ∑ i = 0 M ∑ j = 0 N I FD (i, j)) 2 - 1$	计算区域中分形维数的“块度”,空隙度越大表明区域同质性越强
角二阶矩(ASM)	$AS M FD = ∑ i = 1 M ∑ j = 1 N p ~ 2 (i, j) lg [p ~ (i, j)]$ 其中: $p ~ (i, j) = 1 2 π σ I FD exp (- (I FD (i, j) - μ FD) 2 2 σ FD 2)$	度量区域的紊乱性,值越小,表明区域越均匀,其中, $p ~ (i, j)$ 为分形维数图像中坐标为 $(i, j)$ 处像素点属于所在类的归一化概率值
熵(ENT)	$EN T FD = - ∑ i = 1 M ∑ j = 1 N p ~ (i, j) lg [p ~ (i, j)]$	度量区域内像素的随机性,值越大表明纹理越复杂
同质性(HOM)	$HO M FD = ∑ i = 1 M ∑ j = 1 N p ~ (i, j) 1 + i - j p ~ (i, j)$	描述区域内像素分布均匀度
偏斜度(S)	$S FD = E (x - μ FD) σ FD 3$	描述区域内像素分形纹理值的不对称性
逆差距(IDM)	$ID M FD = ∑ i = 1 M ∑ j = 1 N 1 1 + (i - j) 2 p ~ (i, j)$	类似于同质性,但是更强调区域内像素之间的差异性

Tab.1 分形纹理特征公式及含义

Fig.1 高空间分辨率震害影像分类流程

Fig.2 T1,T2原始影像、MS分割影像以及对应的ROI

Fig.3 T1和T2影像分类结果对比

Tab.2 SVM方法对T1影像的混淆矩阵和精度评价

Tab.3 本文算法对T1影像的混淆矩阵和精度评价

Tab.4 SVM方法对T2影像的混淆矩阵和精度评价

Tab.5 本文算法对T2影像的混淆矩阵和精度评价

[1]	Dong L, Shan J . A comprehensive review of earthquake-induced building damage detection with remote sensing techniques[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2013,84:85-99.
[2]	Vu T T .Rapid disaster damage estimation[C]//International Archives of the Photogrammetry,Remote Sensing and Spatial Information Sciences, 2012, xxxix-b8:65-69.
[3]	Liu J H, Shan X J, Yin J Y . Automatic recognition of damaged town buildings caused by earthquake using remote sensing information:Taking the 2001 Bhuj,India,earthquake and the 1976 Tangshan,China,earthquake as examples[J]. Acta Seismologica Sinica, 2004,17(6):686-696.
[4]	Li X, Yang W, Ao T , et al. An improved approach of information extraction for earthquake-damaged buildings using high-resolution imagery[J]. Journal of Earthquake and Tsunami, 2012,5(4):389-399.
[5]	Fukuoka T , Koshimura S.Quantitative analysis of tsunami debris by object-based image classification of the aerial photo and satellite image[J]. Journal of Japan Society of Civil Engineers Ser B2, 2012, 68:I_371-I_375.
[6]	Vatsavai R, Tuttle M, Bhaduri B , et al. Rapid damage assessment using high-resolution remote sensing imagery:Tools and techniques [C]//Proceedings of the IEEE International Geoscience and Remote Sensing Symposium.IEEE, 2011: 1445-1448.
[7]	Cheng J, Li L, Luo B , et al. High-resolution remote sensing image segmentation based on improved RIU-LBP and SRM[J]. Eurasip Journal on Wireless Communications and Networking, 2013,2013(1):1-12.
[8]	Kuldeep, Garg P K . Texture based information extraction from high resolution images using object based classification approach [C]//Proceedings of the International Workshop on Earth Observation and Remote Sensing Applications, 2014: 299-303.
[9]	Backes A R, Casanova D, Bruno O M . Color texture analysis based on fractal descriptors[J]. Pattern Recognition, 2012,45(5):1984-1992.
[10]	Florindo J B, Bruno O M . Texture analysis by multi-resolution fractal descriptors[J]. Expert Systems with Applications, 2013,40(10):4022-4028.
[11]	Pentland A P . Fractal-based description of natural scenes[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1983,6(6):661-674.
[12]	魏向辉 . 基于分形理论的遥感影像土地利用分类研究[D]. 北京:中国地质大学(北京), 2016.
	Wei X H . Study on Classification of Land Use with Remote Sensing Image Based on Fractal Theory[D]. Beijing:China University of Geosciences(Beijing), 2016.
[13]	杨骇, 韩留生, 陈水森 , 等. 一种基于城市水体指数与分形几何算法的OLI遥感影像水体提取方法[J].测绘通报, 2018(4):44-49.
	Yang H, Han L S, Chen S S , et al. Extracting water information from OLI remote sensing images based on city water index and fractal geometry[J].Bulletin of Surveying and Mapping, 2018(4):44-49.
[14]	Filippi A, Klein A G, Dobreva I , et al. Self-Organizing Map-Based Applications in Remote Sensing[M]. INTECH Open Access Publisher, 2010.
[15]	Jain D K, Dubey S B, Choubey R K , et al. An approach for hyperspectral image classification by optimizing SVM using self organizing map[J]. Journal of Computational Science, 2017,25:252-259.
[16]	Cracknell M J, Reading A M, Caritat P D . Multiple influences on regolith characteristics from continental-scale geophysical and mineralogical remote sensing data using self-organizing maps[J]. Remote Sensing of Environment, 2015,165:86-99.
[17]	Ilc N, Dobnikar A . Generation of a clustering ensemble based on a gravitational self-organising map[J]. Neurocomputing, 2012,96(1):47-56.
[18]	Tao W, Jin H, Zhang Y . Color image segmentation based on mean shift and normalized cuts[J]. IEEE Transactions on Cybernetics, 2007,37(5):1382-1389.
[19]	余国斌, 陈爱斌, 孙华 , 等. 基于Mean Shift的高分辨率遥感影像分割方法[J]. 中南林业科技大学学报, 2012,32(10):168-172.
	Yu G B, Chen A B, Sun H , et al. High-resolution remote sensing image segmentation approach based on mean shift[J]. Journal of Central South University of Forestry and Technology, 2012,32(10):168-172.
[20]	曹帆之, 朱述龙, 朱宝山 , 等. 均值漂移与卡尔曼滤波相结合的遥感影像道路中心线追踪算法[J]. 测绘学报, 2016,45(2):205-212.
	Can F Z, Zhu S L, Zhu B S , et al. Tracking road centerlines from remotely sensed imagery using mean shift and Kalman filtering[J]. Acta Geodaetica et Cartographica Sinica, 2016,45(2):205-212.
[21]	Ming D P, Ci T Y, Cai H , et al. Semivariogram-based spatial bandwidth selection for remote sensing image segmentation with mean-shift algorithm[J]. IEEE Geoscience and Remote Sensing Letters, 2012,9(5):813-817.
[22]	Mandelbrot B B . Self-affine fractals and fractal dimension[J]. Phys Scripta, 1985,32(4):257-260.
[23]	Florindo J B, Backes A R, Castro M D , et al. A comparative study on multiscale fractal dimension descriptors[J]. Pattern Recognition Letters, 2012,33(6):798-806.
[24]	Julesz B , Experiments in the visual perception of texture[J]. Scientific American, 1975,232(4):34-43.
[25]	Azimi J, Fern X . Adaptive cluster ensemble selection[J]. Statistical Analysis and Data Mining the Asa Data Science Journal, 2009,1(3):128-141.
[26]	Fern X Z, Brodley C E . Solving cluster ensemble problems by bipartite graph partitioning [C]//Proceedings of the International Conference on Machine Learning, 2004: 36.
[27]	Fern X Z, Wei L . Cluster ensemble selection[J]. Statistical Analysis and Data Mining, 2008,1(3):128-141.
[28]	Vega-Pons S, Correa-Morris J, Ruiz-Shulcloper J . Weighted Cluster Ensemble Using a Kernel Consensus Function[M]. Springer Berlin Heidelberg, 2008.
[29]	罗会兰, 危辉 . 一致性函数研究[J]. 计算机工程与应用, 2009,45(17):1-4. doi: 10.3778/j.issn.1002-8331.2009.17.001
	Luo H L, Wei H . Study on consensus function based on clustering algorithms of categorical data. Computer Engineering and Applications, 2009,45(17):1-4.
[30]	Strehl A, Ghosh J . Relationship-Based Clustering and Visualization for High-Dimensional Data Mining[M]. INFORMS, 2003.

[1]	范莹琳, 娄德波, 张长青, 魏英娟, 贾福东. 基于面向对象的铁尾矿信息提取技术研究——以迁西地区北京二号遥感影像为例[J]. 自然资源遥感, 2021, 33(4): 153-161.
[2]	张春森, 吴蓉蓉, 李国君, 崔卫红, 冯晨轶. 面向对象的高空间分辨率遥感影像箱线图变化检测方法[J]. 国土资源遥感, 2020, 32(2): 19-25.
[3]	夏既胜, 马梦莹, 符钟壬. 基于GF-2遥感影像的机械性破损面提取方法[J]. 国土资源遥感, 2020, 32(2): 26-32.
[4]	杨丽萍, 马孟, 谢巍, 潘雪萍. 干旱区Landsat8全色与多光谱数据融合算法评价[J]. 国土资源遥感, 2019, 31(4): 11-19.
[5]	刘玉锋, 潘英, 李虎. 基于高空间分辨率遥感数据的天山云杉树冠信息提取研究[J]. 国土资源遥感, 2019, 31(4): 112-119.
[6]	姚丙秀, 黄亮, 许艳松. 一种结合超像素和图论的高空间分辨率遥感影像分割方法[J]. 国土资源遥感, 2019, 31(3): 72-79.
[7]	周阳, 张云生, 陈斯飏, 邹峥嵘, 朱耀晨, 赵芮雪. 基于DCNN特征的建筑物震害损毁区域检测[J]. 国土资源遥感, 2019, 31(2): 44-50.
[8]	陈震, 张耘实, 章远钰, 桑玲玲. 高标准农田建后遥感监测方法[J]. 国土资源遥感, 2019, 31(2): 125-130.
[9]	滑永春, 李增元, 高志海, 郭中. 基于GF-2民勤县白刺包提取技术[J]. 国土资源遥感, 2017, 29(1): 71-77.
[10]	杨青山, 张华. 融合像元形状和光谱信息的高分遥感图像分类新方法[J]. 国土资源遥感, 2016, 28(4): 64-70.
[11]	李雪冬, 张洪岩, 杨广斌, 李晓松, 周越. 辅以地貌类型的喀斯特地区土地利用信息提取[J]. 国土资源遥感, 2016, 28(3): 138-145.
[12]	焦其松, 张景发, 蒋洪波, 宿渊源, 王旭. 基于TLS技术的典型建筑物震害信息三维建模分析——以彭州市白鹿中学为例[J]. 国土资源遥感, 2016, 28(1): 166-171.
[13]	李向辉, 陈一祥, 王海斌, 张恩兵, 秦昆. 基于Gabor滤波和局部特征点密度的居民区提取[J]. 国土资源遥感, 2015, 27(3): 59-64.
[14]	赵志明, 周小成, 付乾坤, 汪小钦. 基于资源三号影像的建筑物高度信息提取方法[J]. 国土资源遥感, 2015, 27(3): 19-24.
[15]	杨德贺, 陈宜金, 杨溪, 吕京国, 张子昕, 张帅. 面向震害信息提取的多源遥感图像自动配准[J]. 国土资源遥感, 2013, 25(3): 56-60.

Viewed

Full text

Abstract

Cited

Shared

Discussed