OBIA与RF结合的龙口市土地利用信息提取方法

doi:10.31497/zrzyxb.20190403

摘要/Abstract

摘要：

为提高中分辨率遥感影像解译精度,本文提出面向对象影像分析（Object Based Image Analysis,OBIA）与随机森林（Random Forest,RF）结合的土地利用信息提取方法。采用Landsat 8 OLI影像,针对不同地物特点,阈值分割和多尺度分割结合创建影像对象,规则集和分类器协同分类,基于Relief F算法分别对光谱特征、纹理特征及所有特征降维筛选特征子集,并与全部特征一起应用RF建模,对龙口市进行土地利用信息提取与比较。结果表明：OBIA与RF结合提取土地利用信息,基于Relief F算法筛选纹理特征,保留完整光谱、几何、空间关系特征构建RF模型,建模错分率为0.0958,分类总体精度和Kappa系数分别为89.37%和0.872,取得较理想结果。该方法可应用于中分辨率遥感影像土地利用信息提取。

关键词: 土地利用信息, 提取方法, 面向对象, Relief F降维, 随机森林, 龙口市

Abstract:

In order to improve the interpretation precision of the medium resolution satellite image, this paper proposed a new extraction approach of land use information combining Object Based Image Analysis (OBIA) with Random Forest (RF). Using the Landsat 8 OLI image and according to the features of all kinds of ground objects, the image objects were created combined with the multi-threshold and multi-resolution segmentation method, and the rule set and classifier were collaboratively used in the image classification. The Relief F algorithm was used to dimensionally reduce the spectral, texture and all feature variables, and to select 3 feature subsets. Then the RF model was conducted with the 3 feature subsets and all feature subset to build 4 models. The 4 models were applied to extract land use information in Longkou city, and the results were compared. The result indicated that the OOB (Out of Bag) misclassification, classification accuracy and Kappa index were 0.0958, 89.37% and 0.872 respectively with the land use information extraction approach combining OBIA with RF, dimension reduction based on the Relief F algorithm only for texture features. This retained the complete spectral, geometric and spatial features, which has a higher accuracy. The approach can be applied to the extraction of land use information with the medium resolution satellite image.

Key words: land use information, the extraction approach, object-based, Relief F algorithm dimensionally reduced, Random Forest, Longkou city

王瑷玲, 张校千, 苏晨晨, 于新洋. OBIA与RF结合的龙口市土地利用信息提取方法[J]. 自然资源学报, 2019, 34(4): 707-717.

WANG Ai-ling, ZHANG Xiao-qian, SU Chen-chen, YU Xin-yang. The extraction approach of land use information combining OBIA with RF in Longkou city[J]. JOURNAL OF NATURAL RESOURCES, 2019, 34(4): 707-717.

图/表 10

图1

图2

表1

表2

图3

表3

图4

表4

表5

表6

参考文献 24

[1]	杜习乐, 吕昌河, 王海荣. 土地利用/覆被变化(LUCC)的环境效应研究进展. 土壤, 2011, 43(3): 350-360.
	[DU X L, LYU C H, WANG H R.Review: Researches on environmental effects of land use/cover change. Soils, 2011, 43(3): 350-360.]
[2]	ROUNSEVELL A, PEDROLI G B M, ERB K H, et al. Challenges for land system science. Land Use Policy, 2012, 29(4): 899-910.
[3]	刘纪远, 匡文慧, 张增祥, 等. 20世纪80年代末以来中国土地利用变化的基本特征与空间格局. 地理学报, 2014, 69(1): 3-14.
	[LIU J Y, KUANG W H, ZHANG Z X, et al.Spatiotemporal characteristics, patterns and causes of land use changes in China since the late 1980s. Acta Geographica Sinica. 2014, 69(1): 3-14.]
[4]	于新洋, 张安定, 侯西勇. 胶东半岛果园TM影像信息的提取决策树方法. 测绘科学, 2012, 37(4): 57-60.
	[YU X Y, ZHANG A D, HOU X Y.Decision tree classification of orchard information extraction from TM imagery in Jiaodong Peninsula of China. Science of Surveying and Mapping, 2012, 37(4): 57-60.]
[5]	王彩艳, 王瑷玲, 王介勇, 等. 基于面向对象的海岸带土地利用信息提取研究. 自然资源学报, 2014, 29(9): 1589-1597.
	[WANG C Y, WANG A L, WANG J Y, et al.Coastal zone land use information extraction based on object-oriented classification method. Journal of Natural Resources, 2014, 29(9): 1589-1597.]
[6]	陈元鹏, 罗明, 彭军还, 等. 基于网格搜索随机森林算法的工矿复垦区土地利用分类. 农业工程学报, 2017, 33(14): 250-257.
	[CHEN Y P, LUO M, PENG J H, et al.Classification of land use in industrial and mining reclamation area based grid-search and random forest classifier. Transactions of the CSAE, 2017, 33(14): 250-257.]
[7]	BREIMAN L.Random forests. Machine Learning, 2001, 45(1): 5-32.
[8]	BARDAN G, JOHN R, VICTOR R G, et al.An evaluation of bagging, boosting, and random forests for land-cover classification in cape cod, Massachusetts, USA. Mapping Sciences & Remote Sensing, 2012, 49(5): 623-643.
[9]	孙雪莲, 舒清态, 欧光龙, 等. 基于随机森林回归模型的思茅松人工林生物量遥感估测. 林业资源管理, 2015, (1): 71-76.
	[SUN X L, SHU Q T, OU G L et al. Remote sensing estimation of the biomass of artificial simao pine forest based on random forest regression. Forest Resources Management, 2015, (1): 71-76.]
[10]	王书玉, 张羽威, 于振华. 基于随机森林的洪河湿地遥感影像分类研究. 测绘与空间地理信息, 2014, 37(4): 83-85.
	[WANG S Y, ZHANG Y W, YU Z H.Classification of Honghe wetland remote sensing image based on random forests. Geomatics & Spatial Information Technology, 2014, 37(4): 83-85.]
[11]	张秀英, 冯学智, 江洪. 面向对象分类的特征空间优化. 遥感学报, 2009, 13(4): 664-677.
	[ZHANG X Y, FENG X Z, JIANG H.Object-oriented classification based on feature space optimization. Journal of Remote Sensing, 2009, 13(4): 664-677.]
[12]	WANG Y, MAKEDON F.Application of Relief-F feature filtering algorithm to selecting informative genes for cancer classification using microarray data//IEEE computational systems bioinformatics conference. IEEE Computer Society, 2004: 497-498.
[13]	肖艳, 姜琦刚, 王斌, 等. 基于Relief F和PSO混合特征选择的面向对象土地利用分类. 农业工程学报, 2016, 32(4): 211-216.
	[XIAO Y, JIANG Q G, WANG B, et al.Object based land-use classification based on hybrid feature selection method of combining Relief F and PSO. Transactions of the CSAE, 2016, 32(4): 211-216.]
[14]	MA L, FU T Y, BLASCHKE T, et al.Evaluation of feature selection methods for object-based land cover mapping of unmanned aerial vehicle imagery using random forest and support vector machine classifiers. ISPRS International Journal of Geo-Information, 2017, 6(2): 51.
[15]	洪波. 高光谱遥感图像信噪比估算方法研究. 北京: 中国科学院大学, 2013.
	[HONG B.Study on methods for SNR estimation of hyperspectral remote sensing images. Beijing: University of Chinese Academy Sciences, 2013.]
[16]	周廷刚, 郭达志. 灰度矢量多波段遥感影像纹理特征及其描述. 西安科技大学学报, 2000, 20(4): 336-338.
	[ZHOU T G, GUO D Z.A study of texture and description for multiband remote sensing image based on the grey vector. Journal of Xi'an University of Science & Technology, 2000, 20(4): 336-338.]
[17]	吴泉源, 侯志华, 于竹洲, 等. 龙口市海岸带土地利用动态变化分析. 地理研究, 2006, 25(5): 921-929.
	[WU Q Y, HOU Z H, YU Z Z, et al.Analysis of the dynamic change of land use in Longkou city's coastal zone based on remote sensing technology. Geographical Research, 2006, 25(5): 921-929.]
[18]	BENZ U C, HOFMANN P, WILLHAUCK G, et al.Multi-resolution, object-oriented fuzzy analysis of remote sensing data for GIS-ready information. ISPRS Journal of Photogrammetry & Remote Sensing, 2004, 58: 239-258.
[19]	常春艳, 赵庚星, 王凌, 等. 滨海光谱混淆区面向对象的土地利用遥感分类. 农业工程学报, 2012, 28(5): 226-231.
	[CHANG C Y, ZHAO G X, WANG L, et al.Land use classification based on RS object-oriented method in coastal spectral confusion region. Transactions of the CSAE, 2012, 28(5): 226-231.]
[20]	WANG Y, MAKEDON F.Application of Relief F feature filtering algorithm to selecting informative genes for cancer classification using microarray data. IEEE Computational Systems Bioinformatics, 2004: 497-498.
[21]	EISAVI V, HOMAYOUNI S, YAZDI A M, et al.Land cover mapping based on random forest classification of multitemporal spectral and thermal images. Environmental Monitoring & Assessment, 2015, 187(5): 1-14.
[22]	张贵花, 王瑞燕, 赵庚星, 等. 基于物候参数和面向对象法的濒海生态脆弱区植被遥感提取. 农业工程学报, 2018, 34(4): 209-216.
	[ZHANG G H, WANG R Y, ZHAO G X, et al.Extraction of vegetation information in coastal ecological vulnerable areas from remote sensing data based on phenology parameters and object-oriented method. Transactions of the CSAE, 2018, 34(4): 209-216.]
[23]	黄昕, 张良培, 李平湘. 融合形状和光谱的高空间分辨率遥感影像分类. 遥感学报, 2007, 11(2): 193-200.
	[HUANG X, ZHANG L P, LI P X.Classification of high spatial resolution remotely sensed imagery based on the fusion of spectral and shape features. Journal of Remote Sensing, 2007, 11(2): 193-200.]
[24]	SEGL K, ROESSNER S, HEIDEN U, et al.Fusion of spectral and shape features for identification of urban surface cover types using reflective and thermal hyperspectral data. ISPRS Journal of Photogrammetry & Remote Sensing, 2003, 58(1): 99-112.

地类	城乡建设用地	耕地	林地	草地	园地	交通用地	未利用地及其他	总计
训练样本	33118	30717	57932	16826	47798	4089	12567	192913
验证样本	16488	15366	28949	8412	22291	2032	6293	93173

特征类别	特征变量及编号	合计/个
光谱特征	1~9：2~6波段均值、MNDWI、NDVI、光谱最大差分Max. Diff 10~20：2~6波段方差、MNDWI方差、NDVI方差、TC 1~TC 3方差 21~23：TC 1~TC 3均值：Brightness、Wetness、Greenness	23
几何特征	1~8：面积、边界指数、紧致度、长宽比、对称性、密度、矩形拟合度、形状指数	8
纹理特征	1~40：GLCM Ang. 2nd moment、GLCM Mean、GLCM Dissimilarity、GLCM Contrast、GLCM Homogeneity、GLCM Entropy、GLCM Correlation、GLCM StdDev (all dir.、0°、45°、90°、135°) 41~60：GLDV Ang. 2nd moment、GLDV Contrast、GLDV Entropy、GLDV Mean (all dir.、0°、45°、90°、135°)	60
空间特征	1~3：5、6 波段和NDVI对象邻域的平均差分（绝对值）	3

参数	Model 1	Model 2	Model 3	Model 4
n_tree	（1,2,4,9,18）	（1,2,4,7,11）	（1,2,4,7,11）	（1,2,4,10,19）
m_try	（135,140,145）	（138,148,158）	（137,144,151）	（140,150,160）

分类模型	算法执行时间	OOB错分率	总体精度/%	Kappa系数
Model 1(SPE_RF)	42 m 34 s	0.1075	84.93	0.818
Model 2(TEX_RF)	39 m 49 s	0.0958	89.37	0.872
Model 3(ALL_RF)	38 m 32 s	0.1026	87.52	0.849
Model 4(ALL)	47 m 21 s	0.1028	87.16	0.845

地物类别	坑塘水面	滩涂	草地	园地	耕地	未利用地及其他	林地		城乡建设用地		交通用地		河流		水库水面
坑塘水面	100	6.34	0	0	0	0	0	0		0		0		0		0	2.01
滩涂	0	61.77	0	0	0	0	0	0		0		0		0		0	2.69
草地	0	1.39	76.44	4.33	3.41	0	0	0		0		0		0		0	7.83
园地	0	0	18.40	89.57	3.73	3.83	2.55	0		0		0		0		0	20.74
耕地	0	0	4.33	1.88	86.45	1.16	0	0.31		0		0		0		0	14.30
未利用地及其他	0	22.61	0	0	6.41	75.10	0.76	2.24		0		0		0		0	5.85
林地	0	1.00	0.31	3.62	0	1.43	96.59	0		0		0		0		0	26.93
城乡建设用地	0	1.52	0	0	0	17.64	0	92.72		8.61		0		0		0	15.34
交通用地	0	4.29	0.52	0.60	0	0.84	0.09	4.73		91.39		0		0		0	1.89
河流	0	0	0	0	0	0	0	0		0		100		0		0	0.85
水库水面	0	1.07	0	0	0	0	0	0		0		0		100		0	1.58
未分类	0	0	0	0	0	0	0	0		0		0		0		0	0
生产者精度	100	61.77	76.44	89.57	86.45	75.10	96.59		92.72		91.39		100		100