多时相耕地覆盖提取和变化分析：一种结合遥感和空间统计的时空上下文方法

doi:10.12082/dqxxkx.2022.210034

摘要/Abstract

摘要：

基于多时相影像的耕地提取和变化分析是有效管理和保护耕地资源的重要手段。然而就多时相耕地的分类提取而言,现有方法对于多时相影像中地物的时空特征表达和时空上下文关系建模存在着局限性,导致耕地的提取精度不佳;其次,对于耕地的变化分析,现有方法往往只关注基于行政单元的耕地面积统计变化,而对耕地变化在空间上的相关性分布特点考虑较少。因此,本文首先提出了一种时空上下文分类方法,综合表达和利用多时相影像中地物的光谱、纹理和空间等特征,建模时空维度上地物间在特征和语义上的上下文关系,来提高耕地覆盖分类的精度;其次,基于耕地覆盖的提取结果,在规则格网和行政区划单元上,采用GIS空间统计方法分析耕地变化的空间相关性特点;最后,以北京市顺义区为例,以2015—2019年的多时相Sentinel-2影像为数据源对本文方法进行验证。结果表明,与常见的2种多时相影像分类方法相比,本文方法在多时相耕地分类上精度最高,平均用户精度和制图精度分别达到91.21%和90.53%,所有类别的总体精度为90.79%。这表明本文方法能精确提取多时相耕地覆盖信息。通过对耕地变化的空间分布特点进行分析,发现2015—2019年顺义区耕地变化存在区域聚集现象,主要呈现为集中减少特点,其中赵全营镇、高丽营镇、木林镇和杨镇地区耕地的聚集性减少较为明显,说明这些地区的耕地侵占和减少问题比较严重。

关键词: 耕地, 多时相数据, 时空特征, 时空上下文, 条件随机场, 影像分类, 空间统计, 变化分析

Abstract:

The extraction and change analysis of cultivated land covers based on multi-temporal images are important means to effectively manage and protect cultivated land resources. However, as far as the classification and extraction of multi-temporal cultivated land covers are concerned, the existing methods have limitations in the comprehensive expression of spatiotemporal features and the accurate modeling of spatiotemporal context relations among geographic objects, leading to the poor extraction accuracy of cultivated land covers. In addition, for the analysis of cultivated land change, the existing methods usually only focus on the statistical areal change of cultivated land cover based on administrative units, while consideration is seldom taken into the spatial correlation distribution characteristics of changes in cultivated land covers. Accordingly, first of all, this paper proposes a multi-temporal spatiotemporal context classification method, which comprehensively expresses and utilizes the multi-temporal spectral, texture, and spatial features of geographic objects, and models the contextual relations of features and semantics among geographic objects in both spatial and temporal dimensions of multi-temporal images, so as to improve the classification accuracy of cultivated land covers. Then, based on the extracted results of cultivated land covers, spatial statistical method of Geographic Information System (GIS) is used to analyze the spatial correlation characteristics of cultivated land changes in regular grids and administrative division units. Finally, Shunyi District of Beijing is taken as the study area while multi-temporal Sentinel-2 images in 2015-2019 are used as the data sources to conduct verification of the proposed method. The results show that, compared with the two existing common multi-temporal classification methods, the proposed method achieves the highest accuracies in the classification of multi-temporal cultivated lands. The average user's accuracy and producer's accuracy reach 91.21% and 90.53%, respectively, while the overall accuracy of all categories is 90.79%, indicating that the proposed method can accurately extract the multi-temporal cultivated land cover information. Furthermore, according to the analysis of the spatial distribution characteristics of cultivated land changes, this study found a phenomenon of regional aggregation of the cultivated land change in Shunyi District from 2015 to 2019, which mainly presents the characteristics of concentrated reduction. The aggregation reduction phenomena of cultivated land covers in Zhaoquanying Town, Gaoliying Town, Mulin Town, and Yang Town are especially obvious, indicating that the problems of cultivated land encroachment and reduction are quite serious in these areas.

Key words: cultivated land, multi-temporal data, spatiotemporal features, spatiotemporal context, conditional random field, image classification, spatial statistics, change analysis

习文强, 杜世宏, 杜守基. 多时相耕地覆盖提取和变化分析：一种结合遥感和空间统计的时空上下文方法[J]. 地球信息科学学报, 2022, 24(2): 310-325.DOI:10.12082/dqxxkx.2022.210034

XI Wenqiang, DU Shihong, DU Shouji. Multi-temporal Cultivated Land Cover Extraction and Change Analysis: A Spatiotemporal Context Method Combining Remote Sensing and Spatial Statistics[J]. Journal of Geo-information Science, 2022, 24(2): 310-325.DOI:10.12082/dqxxkx.2022.210034

图/表 20

图1

表1

图2

图3

表2

图4

表3

图5

表4

图6

表5

表6

图7

表7

图8

图9

图10

图11

图12

图13

参考文献 26

[1]	Smil V. China's energy and resource uses: continuity and change[J]. The China Quarterly, 1998, 156:935-951. doi: 10.1017/S0305741000051407
[2]	毕于运. 中国耕地[M]. 北京: 中国农业科技出版社, 1995:1-9.
	[ Bi Y Y. China cultivated land[M]. Beijing: China Agricultural Science and Technology Publishing, 1995:1-9. ]
[3]	赵其国, 周生路, 吴绍华, 等. 中国耕地资源变化及其可持续利用与保护对策[J]. 土壤学报, 2006, 43(4):662-672.
	[ Zhao Q G, Zhou S L, Wu S H, et al. Cultivated land resources and strategies for its sustainable utilization and protection in China[J]. Acta Pedologica Sinica, 2006, 43(4):662-672. ]
[4]	张国平, 刘纪远, 张增祥. 近10年来中国耕地资源的时空变化分析[J]. 地理学报, 2003, 58(3):323-332.
	[ Zhang G P, Liu J Y, Zhang Z X. Spatial-temporal changes of cropland in China for the past 10 years based on remote sensing[J]. Acta Geographica Sinica, 2003, 58(3):323-332. ]
[5]	Lillesand T, Kiefer R W, Chipman J. Remote sensing and image interpretation[M]. John Wiley & Sons, 2015.
[6]	Giuliani G, Dao H, De Bono A, et al. Live Monitoring of Earth Surface (LiMES): A framework for monitoring environmental changes from Earth Observations[J]. Remote Sensing of Environment, 2017, 202:222-233. doi: 10.1016/j.rse.2017.05.040
[7]	Xi W, Du S. What information is important? A spatiotemporal inference for classification of satellite image time series[C]. IGARSS 2019-2019 IEEE International Geoscience and Remote Sensing Symposium. IEEE, 2019:2415-2418.
[8]	魏鹏飞, 徐新刚, 杨贵军, 等. 基于多时相影像植被指数变化特征的作物遥感分类[J]. 中国农业科技导报, 2019, 21(2):54-61.
	[ Wei P F, Xu X G, Yang G J, et al. Crop remote sensing classification based on vegetation index change characteristics of multi temporal images[J]. Chinese Journal of Agricultural Science and Technology, 2019, 21(2):54-61. ]
[9]	Slagter B, Tsendbazar N E, Vollrath A, et al. Mapping wetland characteristics using temporally dense Sentinel-1 and Sentinel-2 data: A case study in the St. Lucia wetlands, South Africa[J]. International Journal of Applied Earth Observation and Geoinformation, 2020, 86:102009. doi: 10.1016/j.jag.2019.102009
[10]	阴海明, 王立辉, 董明霞, 等. 基于多时相Sentinel-2遥感影像的江汉平原夏收作物提取方法[J]. 福建农林大学学报(自然科学版), 2021, 50(1):16-22.
	[ Ying H M, Wang L H, Dong M X, et al. Extraction method of summer harvest crops in Jianghan plain based on multi-temporal Sentinel-2 remote sensing images[J]. Journal of Fujian Agriculture and Forestry University (Natural Science Edition), 2021, 50(01):16-22. ]
[11]	Peña M A, Brenning A. Assessing fruit-tree crop classification from Landsat-8 time series for the Maipo Valley, Chile[J]. Remote Sensing of Environment, 2015, 171:234-244. doi: 10.1016/j.rse.2015.10.029
[12]	Xi W, Du S, Wang Y C, et al. A spatiotemporal cube model for analyzing satellite image time series: Application to land-cover mapping and change detection[J]. Remote Sensing of Environment, 2019, 231:111212. doi: 10.1016/j.rse.2019.111212
[13]	Blaschke T. Object based image analysis for remote sensing[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2010, 65(1):2-16. doi: 10.1016/j.isprsjprs.2009.06.004
[14]	Petitjean F, Kurtz C, Passat N, et al. Spatio-temporal reasoning for the classification of satellite image time series[J]. Pattern Recognition Letters, 2012, 33(13):1805-1815. doi: 10.1016/j.patrec.2012.06.009
[15]	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. doi: 10.1109/TGRS.2014.2326886
[16]	Wehmann A, Liu D. A spatial-temporal contextual Markovian kernel method for multi-temporal land cover mapping[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2015, 107:77-89. doi: 10.1016/j.isprsjprs.2015.04.009
[17]	Liu D S, Cai S S. A spatial-temporal modeling approach to reconstructing land-cover change trajectories from multi-temporal satellite imagery[J]. Annals of the Association of American Geographers, 2012, 102(6):1329-1347. doi: 10.1080/00045608.2011.596357
[18]	Giri C, Muhlhausen J. Mangrove forest distributions and dynamics in Madagascar (1975—2005)[J]. Sensors (Basel, Switzerland), 2008, 8(4):2104-2117. doi: 10.3390/s8042104
[19]	柏林川, 武兰芳, 宋小青. 1995—2010年山东省粮食单产变化空间分异及均衡增产潜力[J]. 地理科学进展, 2013, 32(8):1257-1265.
	[ Bo L C, Wu L F, Song X Q. Spatial difference of grain yield changes during 1995-2010 and balanced potential output to increase in Shandong Province[J]. Progress in Geography, 2013, 32(8):1257-1265. ]
[20]	Baatz M. Multi resolution segmentation: An optimum approach for high quality multi scale image segmentation[C]. Beutrage zum AGIT-Symposium. Salzburg, Heidelberg, 2000:12-23.
[21]	Ding S F, Su C Y, Yu J Z. An optimizing BP neural network algorithm based on genetic algorithm[J]. Artificial Intelligence Review, 2011, 36(2):153-162. doi: 10.1007/s10462-011-9208-z
[22]	Shotton J, Winn J, Rother C, et al. Texton Boost for image understanding: Multi-class object recognition and segmentation by jointly modeling texture, layout, and context[J]. International Journal of Computer Vision, 2009, 81(1):2-23. doi: 10.1007/s11263-007-0109-1
[23]	Boykov Y, Veksler O, Zabih R. Fast approximate energy minimization via graph cuts[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2001, 23(11):1222-1239. doi: 10.1109/34.969114
[24]	Fotheringham A S, Brunsdon C, Charlton M. Quantitative geography: Perspectives on spatial data analysis[M]. London: SAGE Publication, 2000.
[25]	Foody G M. Classification accuracy comparison: Hypojournal tests and the use of confidence intervals in evaluations of difference, equivalence and non-inferiority[J]. Remote Sensing of Environment, 2009, 113(8):1658-1663. doi: 10.1016/j.rse.2009.03.014
[26]	Suhet. Sentinel-2 User Handbook[EB/OL]. https://sentinel.esa.int/documents/247904/685211/Sentinel-2_User_H and book, 2013-9-1.

类型	维度	描述
光谱特征	1—8	R, G, B, NIR, SWIR, NDVI, EVI, LSWI
纹理特征	9—12	GLCM-Homogeneity
	13—16	GLCM-Correlation
	17—20	GLCM-Energy
	21—24	GLCM-Entropy
	25—28	GLCM-Contrast
空间特征	29—36	面积、边界长度、长宽比、矩形度、椭圆度、紧凑度、光滑度和形状指数

t	t+1
t	耕地	林地	草地	裸土	建筑物	道路	水体
耕地	0.90	0.04	0.04	0.10	0.15	0.01	0.01
林地	0.02	0.90	0.02	0.01	0.01	0.01	0.01
草地	0.02	0.02	0.90	0.02	0.01	0.01	0.01
裸土	0.02	0.02	0.02	0.90	0.10	0.02	0.01
建筑物	0.01	0.01	0.01	0.08	0.90	0.05	0.05
道路	0.01	0.01	0.01	0.02	0.01	0.90	0.01
水体	0.01	0.01	0.01	0.01	0.01	0.01	0.90

参数	数值
空间分辨率	10 m
光谱波段	B2 “Blue” (490 nm) B3 “Green” (560 nm) B4 “Red” (665 nm) B8 “Near-Infrared” (842 nm) B6 “Short Wave Infrared 1” (1610 nm)
成像时间	2015-08-13, 2016-08-21, 2017-08-06, 2018-08-16, 2019-08-18

样本/像素	类别
样本/像素	耕地	林地	草地	裸土	建成区	道路	水体
训练	1 382 063	478 291	165 725	115 176	1 359 133	448 467	171 082
测试	690 474	238 686	80 637	55 349	678 865	224 026	83 813

类别		分类方法
类别		NOC	PPC	ST-PC ST-CRF
用户精度	耕地	81.27	88.14	91.21
	林地	80.01	87.96	90.03
	草地	80.36	87.32	88.92
	裸土	87.19	89.25	90.85
	建成区	88.82	84.16	91.84
	道路	89.73	83.58	90.41
	水体	96.79	95.06	96.62
制图精度	耕地	80.83	87.75	90.53
	林地	80.94	87.24	89.72
	草地	81.59	88.38	90.98
	裸土	86.93	88.62	91.06
	建成区	88.95	84.76	91.17
	道路	88.57	84.19	90.64
	水体	96.02	95.86	96.93