基于深度学习的寒旱区多时序影像土地利用及变化监测——以新疆莫索湾垦区为例

doi:10.12118/j.issn.1000–6060.2021.06.20

干旱区地理 ›› 2021, Vol. 44 ›› Issue (6): 1717-1728.doi: 10.12118/j.issn.1000–6060.2021.06.20

基于深度学习的寒旱区多时序影像土地利用及变化监测——以新疆莫索湾垦区为例

袁盼丽^1,^2,³(),汪传建⁴,赵庆展^1,^2,³(),王学文^1,^2,³,任媛媛^1,²,杨启原^1,²

1. 石河子大学信息科学与技术学院,新疆石河子 832003
2. 兵团空间信息工程技术研究中心,新疆石河子 832003
3. 兵团工业技术研究院,新疆石河子 832003
4. 安徽大学互联网学院,安徽合肥 230039

收稿日期:2020-12-08 修回日期:2021-04-27 出版日期:2021-11-25 发布日期:2021-12-03
通讯作者: 赵庆展
作者简介:袁盼丽（1995-）,女,硕士研究生,主要从事机器学习、遥感图像处理研究. E-mail: yuanpanli@stu.shzu.edu.cn
基金资助:
国家重点研发计划项目(2017YFB0504203);新疆生产建设兵团科技计划项目(2017DB005)

Dynamic monitoring of land-use/land-cover change in cold and arid region based on deep learning: A case study of Mosuowan reclamation area in Xinjiang

YUAN Panli^1,^2,³(),WANG Chuanjian⁴,ZHAO Qingzhan^1,^2,³(),WANG Xuewen^1,^2,³,REN Yuanyuan^1,²,YANG Qiyuan^1,²

1. School of Information Science and Technology, Shihezi University, Shihezi 832003, Xinjiang, China
2. Research Center for Space Information Engineering Technology of XPCC, Shihezi 832003, Xinjiang, China
3. XPCC Industrial Technology Research Institute, Shihezi 832003, Xinjiang, China
4. Internet College of Anhui University, Hefei 230039, Anhui, China

Received:2020-12-08 Revised:2021-04-27 Online:2021-11-25 Published:2021-12-03
Contact: Qingzhan ZHAO

摘要/Abstract

摘要：

针对生态环境脆弱的寒旱区开展地物要素提取以及土地覆盖变化监测研究,对农业规划、城乡建设、生态环境监测与保护等具有重要意义。借助2015—2019年新疆莫索湾垦区Landsat-8影像构建数据集,对比3种传统方法：最大似然分类（Maximum likelihood classification,MLC）、支持向量机（Support vector machine,SVM）和随机森林（Random forest,RF）及5种语义分割模型：DeepLabv3+（Xception）、DeepLabv3+（MobileNet）、SegNet（ResNet50）、U-Net（MobileNet）和PSPNet（MobileNet）,选取最优自动化地物提取模型对研究区1998—2020年农用地、建筑用地、水体和荒漠4种地物要素进行分类,并运用土地利用转移矩阵和动态度进行定量动态变化分析。结果表明：DeepLabv3+（Xception）模型可以实现更准确、更高效的地物提取,总体精确度（OA）、Kappa系数和F1值分别为96.06%、0.96和0.86,其中所选模型的平均交并比（MIoU）较其他模型提升0.03~0.39。近23 a,莫索湾垦区的荒漠、农用地和建筑用地三者的土地结构转化较为明显,荒漠总面积减少15.00%,农用地总面积增加12.68%,建筑用地总面积增加2.53%,水体面积变化较为平稳。地物类型总体转变方向为荒漠向农用地转化、农用地向建筑用地转化。该研究可为深度学习技术应用于中分辨率遥感卫星影像领域中实现土地利用及变化动态监测提供参考。

关键词: 遥感影像, 长时间序列, 深度学习, DeepLabv3+, 地物分类, 动态监测

Abstract:

In this study, we conducted researchon the extraction of ground elements and the dynamic monitoring of land cover change in cold and arid areas with fragile ecological environments. The Mosuowan reclamation area in Xinjiang, China was selected as the target area, and the Landsat series satellite images from 2015 to 2019 were cut into subimages with sizes of 416×416 px. Specifically, the training set, validation set and testset are 306 subimages, 204 subimages and 120 subimages respectively. Three traditional methods were then evaluated: MLC,SVM, and RF. Five semantic segmentation models were also evaluated: DeepLabv3+(Xception), DeepLabv3+ (MobileNet), SegNet (ResNet50), U-Net (MobileNet), and PSPNet (MobileNet). In the evaluation experiments, DeepLabv3+ (Xception) was found to achieve the optimal segmentation effect and multiscale feature fusion using fewer parameters. The overall accuracy, Kappa coefficient, precision, recall, F1-score, and MIoU were 96.06%, 0.96, 87.69%, 83.78%, 0.86, and 0.77, respectively. The MIoU of the DeepLabv3+(Xception) model was significantly better than those of the other four models, improving from 0.03 to 0.39. On the basis of the land use classification results of the long-term timeseries remote sensing data from 1998 to 2020, we analyze spatial structure change in land use and the associated driving factors. Over the past 23 years, the total areas of desert, agricultural land, and construction land have been reduced by 15.00%, 12.68%, 2.53%, respectively. At the same time, the amount of water area has remained relatively stable. The overall transformation direction of land use is from desert to agricultural land and then from agricultural land to construction land. It can be seen from the results that the desertification control was effective, and urbanization rapidly developed. Consequently, this study can provide a reference for the application of deep learning in the field of medium-resolution remote sensing images, which can be used to realize the dynamic monitoring of land use and change.

Key words: remote sensing image, long time-series, deep learning, DeepLabv3+, feature classification, dynamic monitoring

袁盼丽,汪传建,赵庆展,王学文,任媛媛,杨启原. 基于深度学习的寒旱区多时序影像土地利用及变化监测——以新疆莫索湾垦区为例[J]. 干旱区地理, 2021, 44(6): 1717-1728.

YUAN Panli,WANG Chuanjian,ZHAO Qingzhan,WANG Xuewen,REN Yuanyuan,YANG Qiyuan. Dynamic monitoring of land-use/land-cover change in cold and arid region based on deep learning: A case study of Mosuowan reclamation area in Xinjiang[J]. Arid Land Geography, 2021, 44(6): 1717-1728.

图/表 13

图1

表1

图2

表2

图3

表3

表4

表5

图4

图5

图6

表6

表7

参考文献 35

[1]	董金玮, 吴文斌, 黄健熙, 等. 农业土地利用遥感信息提取的研究进展与展望[J]. 地球信息科学报, 2020, 22(4):772-783.
	[ Dong Jinwei, Wu Wenbin, Huang Jianxi, et al. State of the art and perspective of agricultural land use remote sensing information extraction[J]. Journal of Geo-information Science, 2020, 22(4):772-783. ]
[2]	Gaur S, Mittal A, Bandyopadhyay A, et al. Spatio-temporal analysis of land use and land cover change: A systematic model inter-comparison driven by integrated modelling techniques[J]. International Journal of Remote Sensing, 2020, 41(23):9229-9255. doi: 10.1080/01431161.2020.1815890
[3]	陈劲松, 韩宇, 陈工, 等. 基于多源遥感信息融合的广东省土地利用分类方法——以雷州半岛为例[J]. 生态学报, 2014, 34(24):7233-7242.
	[ Chen Jinsong, Han Yu, Chen Gong, et al. Land utilization mapping in Guangdong Province based on integration of optical and SAR remote sensing data[J]. Acta Ecologica Sinica, 2014, 34(24):7233-7242. ]
[4]	李路, 孙桂丽, 陆海燕, 等. 喀什地区生态脆弱性时空变化及驱动力分析[J]. 干旱区地理, 2021, 44(1):277-288.
	[ Li Lu, Sun Guili, Lu Haiyan, et al. Spatial-temporal variation and driving forces of ecological vulnerability in Kashi Prefecture[J]. Arid Land Geography, 2021, 44(1):277-288. ]
[5]	Huang B, Zhao B, Song Y M. Urban land-use mapping using a deep convolutional neural network with high spatial resolution multispectral remote sensing imagery[J]. Remote Sensing of Environment, 2018, 214:73-86. doi: 10.1016/j.rse.2018.04.050
[6]	宋德娟, 张承明, 杨晓霞, 等. 高分二号遥感影像提取冬小麦空间分布[J]. 遥感学报, 2018, 24(5):596-608.
	[ Song Dejuan, Zhang Chengming, Yang Xiaoxia, et al. Extracting winter wheat spatial distribution information from GF-2 image[J]. Journal of Remote Sensing, 2018, 24(5):596-608. ]
[7]	Noi P T, Kappas M. Comparison of random forest, k-nearest neighbor, and support vector machine classifiers for land cover classification using Sentinel-2 imagery[J]. Sensors, 2018, 18(1):7233-7242.
[8]	Son S, Park S, Lee S, et al. An assessment of support sector machine for land cover classification over South Korea[J]. Earth Resources and Environmental Remote Sensing/GIS Applications X, 2019, 19(10):2308, doi: 10.1117/12.2533045. doi: 10.1117/12.2533045
[9]	Zhang C, Yue P, Tapete D, et al. A multi-level context-guided classification method with object-based convolutional neural network for land cover classification using very high resolution remote sensing images[J]. International Journal of Applied Earth Observation and Geoinformation, 2020, 88:102086, doi: 10.1016/j.jag.2020.102086. doi: 10.1016/j.jag.2020.102086
[10]	孙坤, 鲁铁定. 顾及多尺度分割参数的FNEA面向对象分类[J]. 测绘通报, 2018(3):43-48.
	[ Sun Kun, Lu Tieding. Research on FNEA object-oriented classification based on multi-scale partition parameters[J]. Bulletin of Surveying and Mapping, 2018(3):43-48. ]
[11]	Zhu X X, Devis T, Mou L C, et al. Deep learning in remote sensing: A comprehensive review and list of resources[J]. IEEE Geoscience and Remote Sensing Magazine, 2017, 5(4):8-36.
[12]	汪传建, 赵庆展, 马永建, 等. 基于卷积神经网络的无人机遥感农作物分类[J]. 农业机械学报, 2019, 50(11):161-168.
	[ Wang Chuanjian, Zhao Qingzhan, Ma Yongjian, et al. Crop identification of drone remote sensing based on convolutional neural network[J]. Transactions of the Chinese Society for Agricultural Machinery, 2019, 50(11):161-168. ]
[13]	Vali Ava, Comai Sara, Matteucci Matto. Deep learning for land use and land cover classification based on hyperspectral and multispectral earth observation data: A review[J]. Remote Sensing, 2020, 12(15):2495, doi: 10.3390/rs12152495. doi: 10.3390/rs12152495
[14]	Long J, Shelhamer E, Darrell T. Fully convolutional networks for semantic segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 39(4):640-651. doi: 10.1109/TPAMI.2016.2572683
[15]	Badrinarayanan V, Kendall A, Cipolla R. SegNet: A deep convolutional encoder-decoder architecture for image segmentation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(12):2481-2495. doi: 10.1109/TPAMI.2016.2644615 pmid: 28060704
[16]	Ronneberger O, Fischer P, Brox T. U-Net: Convolutional networks for biomedical image segmentation [C]//Maier-Hein, Fritzsche K, Lehmann T, et al. Informatik aktuell. Heidelberg: Springer, 2015: 234-241.
[17]	Zhao H, Shi J, Qi X, et al. Pyramid scene parsing network [C]//Proceedings of 2017 IEEE Conference on Computer Vision and Pattern Recognition. USA: IEEE, 2017: 6230-6239.
[18]	Chen L C, Papandreou G, Kokkinos I, et al. DeepLab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected CRFs[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(4):834-848. doi: 10.1109/TPAMI.2017.2699184
[19]	Chen L C, Papandreou G, Schroff F, et al. Rethinking atrous convolution for semantic image segmentation[J]. CoRR, 2017, doi: abs/1706.05587. doi: abs/1706.05587
[20]	Chen L C, Zhu Y K, Papandreou G, et al. Encoder-decoder with atrous separable convolution for semantic image segmentation [C]//Ferrari V, Hebert M, Sminchisescu C, et al. Computer Vision-ECCV 2018. Lecture Notes in Computer Science. Munich: Springer, 2018: 833-851.
[21]	许慧敏, 齐华, 南轲, 等. 结合nDSM的高分辨率遥感影像深度学习分类方法[J]. 测绘通报, 2019(8):63-67.
	[ Xu Huimin, Qi Hua, Nan Ke, et al. High-resolution remote sensing image classification by combining deep learning with nDSM[J]. Bulletin of Surveying and Mapping, 2019(8):63-67. ]
[22]	杨建宇, 周振旭, 杜贞容, 等. 基于SegNet语义模型的高分辨率遥感影像农村建设用地提取[J]. 农业工程学报, 2019, 35(5):259-266.
	[ Yang Jianyu, Zhou Zhenxu, Du Zhenrong, et al. Rural construction land extraction from high spatial resolution remote sensing image based on SegNet semantic segmentation model[J]. Transactions of the Chinese Society of Agricultural Engineering, 2019, 35(5):259-266. ]
[23]	刘文雅, 岳安志, 季珏, 等. 基于DeepLabv3+语义分割模型的GF-2影像城市绿地提取[J]. 国土资源遥感, 2020, 32(2):124-133.
	[ Liu Wenyan, Yue Anzhi, Ji Jue, et al. Urban green space extraction from GF-2 remote sensing image based on DeepLabv3+ semantic segmentation model[J]. Remote Sensing for Land & Resources, 2020, 32(2):124-133. ]
[24]	马新萍, 韩申山, 王磊, 等. 大西安地区土地利用类型时空演变分析[J]. 干旱区地理, 2020, 43(2):499-507.
	[ Ma Xinping, Han Shenshan, Wang Lei, et al. Spatial and temporal evolution of land use types in the greater Xi’an area[J]. Arid Land Geography, 2020, 43(2):499-507. ]
[25]	田艳君, 石莹, 帅艳民, 等. 基于遥感时序特征的地表覆被信息提取[J]. 干旱区地理, 2021, 44(2):450-459.
	[ Tian Yanjun, Shi Ying, Shuai Yanmin, et al. Land cover information retrieval from temporal features based remote sensing images[J]. Arid Land Geography, 2021, 44(2):450-459. ]
[26]	白燕英, 高聚林, 张宝林. 基于Landsat 8影像时间序列NDVI的作物种植结构提取[J]. 干旱区地理, 2019, 42(4):893-901.
	[ Bai Yanying, Gao Julin, Zhang Baolin. Extraction of crop planting structure based on time-series NDVI of Landsat 8 images[J]. Arid Land Geography, 2019, 42(4):893-901. ]
[27]	顾炼. 基于深度学习的遥感图像建筑物检测及其变化检测研究[D]. 杭州: 浙江工商大学, 2018.
	[ Gu Lian. Detection for buildings and their changes in remote sensing images based on deep learning[D]. Hangzhou: Zhejiang Gongshang University, 2018. ]
[28]	Venugopal N. Automatic semantic segmentation with deeplab dilated learning network for change detection in remote sensing images[J]. Neural Processing Letters, 2020, 51(3):2355-2377. doi: 10.1007/s11063-019-10174-x
[29]	Liu R Y, Kuffer M, Persello C. The temporal dynamics of slums employing a CNN-based change detection approach[J]. Remote Sensing, 2019, 11(23):2844, doi: 10.3390/rs11232844. doi: 10.3390/rs11232844
[30]	季顺平, 田思琦, 张驰. 利用全空洞卷积神经元网络进行城市土地覆盖分类与变化检测[J]. 武汉大学学报(信息科学版), 2020, 45(2):233-241.
	[ Ji Shunping, Tian Siqi, Zhang Chi. Urban land cover classification and change detection using fully atrous convolutional neural network[J]. Geomatics and Information Science of Wuhan University, 2020, 45(2):233-241. ]
[31]	邓铭江. 天山北坡经济带“三生空间”发展格局与智能水网体系建设[J]. 干旱区地理, 2020, 43(5):1155-1168.
	[ Deng Mingjiang. Development pattern of production-living-ecological spaces and construction of a smart water network system for the economic belt on the north slope of the Tianshan Mountains[J]. Arid Land Geography, 2020, 43(5):1155-1168. ]
[32]	韩会然, 杨成凤, 宋金平. 北京市土地利用变化特征及驱动机制[J]. 经济地理, 2015(5):148-154.
	[ Han Huiran, Yang Chengfeng, Song Jinping. The spatial-temporal characteristic of land use change in Beijing and its driving mechanism[J]. Economic Geography, 2015(5):148-154. ]
[33]	严彩虹, 周龙, 唐震, 等. 莫索湾垦区冬枣大棚温度特征变化及日最低气温预报研究[J]. 浙江农业科学, 2018, 59(3):445-448.
	[ Yan Caihong, Zhou Long, Tang Zhen, et al. Study on temperature characteristic change and daily minimum temperature forecast of dongzao in Mosuowan reclamation area[J]. Journal of Zhejiang Agricultural Sciences, 2018, 59(3):445-448. ]
[34]	陈曦, 常存, 包安明, 等. 改革开放40 a来新疆土地覆被变化的空间格局与特征[J]. 干旱区地理, 2020, 43(1):1-11.
	[ Chen Xi, Chang Cun, Bao Anming, et al. Spatial pattern and characteristics of land cover change in Xinjiang since past 40 years of the economic reform and opening up[J]. Arid Land Geography, 2020, 43(1):1-11. ]
[35]	郭飞, 陈万山, 吴雪勤, 等. 新疆生产建设兵团第八师石河子市统计年鉴—2019[M]. 北京: 中国统计出版社, 2019: 49-52.
	[ Guo Fei, Chen Wanshan, Wu Xueqin, et al. Statistical yearbook of Division 8 of the Xinjiang Production and Construction Corps: 2019[M]. Beijing: China Statistical Publishing House, 2019: 49-52. ]

传感器类型	发射时间(年-月-日)	空间分辨率/m	时间分辨率/d	年份	云量范围/%
Landsat-5/TM	1984-03-01	30	16	1998,2006—2011	0.00~3.00
Landsat-7/ETM+	1999-04-15	30	16	1999—2005	0.00~6.45
Landsat-8/OLI	2013-02-11	30	16	2013—2020	0.00~2.39

地物类型	包含类型	颜色编码 (Red,Green,Blue)
建筑用地	城镇、村庄、工业建筑用地等	(1,1,1)
农用地	耕地(旱地、水田、光果园等)、林地(灌木林地、有林地等)、草地(天然牧草地、其他草地)等	(2,2,2)
水体	滩涂、坑塘、水库、湿地等	(3,3,3)
荒漠	沙漠、未利用土地等	(4,4,4)
背景	研究区范围以外的区域	(0,0,0)

模型	评价指标
模型	OA/%	Kappa	precision/%	recall/%	F1值	MIoU
MLC	62.73	0.42	52.78	71.17	0.51	0.38
SVM	68.32	0.48	52.12	73.18	0.54	0.41
RF	76.35	0.59	75.59	85.69	0.75	0.64
DeepLabv3+ (Xception)	96.06	0.96	87.69	83.78	0.86	0.77
DeepLabv3+ (MobileNet)	94.77	0.94	74.59	75.82	0.75	0.62
SegNet (ResNet50)	95.47	0.95	81.94	81.04	0.81	0.71
U-Net (MobileNet)	95.83	0.95	85.23	83.13	0.83	0.74
PSPNet (MobileNet)	95.52	0.90	85.34	81.08	0.83	0.74

地物类型	指标	语义分割模型
地物类型	指标	DeepLabv3+ (Xception)	DeepLabv3+ (MobileNet)	SegNet (ResNet50)	U-Net (MobileNet)	PSPNet (MobileNet)
农用地	precision/%	91.25	89.02	89.46	91.68	89.27
	recall/%	91.16	88.82	88.66	89.31	86.85
	F1值	0.91	0.88	0.89	0.90	0.87
建筑用地	precision/%	81.12	66.73	78.65	80.79	75.83
	recall/%	70.36	47.95	63.46	69.07	71.68
	F1值	0.74	0.53	0.69	0.74	0.73
水体	precision/%	99.24	95.58	97.80	97.76	96.31
	recall/%	90.19	96.37	95.37	95.21	96.77
	F1值	0.94	0.96	0.97	0.96	0.97
荒漠	precision/%	81.72	80.17	77.16	77.02	79.55
	recall/%	77.64	67.50	74.66	78.30	70.27
	F1值	0.78	0.70	0.75	0.76	0.73

日期（年-月）	评价指标
日期（年-月）	OA/%	Kappa
1998-08	96.79	0.95
2020-08	97.49	0.96
实验数据	96.06	0.96

基于深度学习的寒旱区多时序影像土地利用及变化监测——以新疆莫索湾垦区为例

Dynamic monitoring of land-use/land-cover change in cold and arid region based on deep learning: A case study of Mosuowan reclamation area in Xinjiang

RichHTML

PDF (PC)

赞

可视化

摘要/Abstract

引用本文

使用本文

图/表 13

参考文献 35

相关文章 3

Metrics

本文评价

推荐阅读 0

2020年地物类型	1998年地物类型
2020年地物类型	农用地	建筑用地	水体	荒漠
农用地	778.40	11.32	1.02	222.39
建筑用地	37.05	11.61	0.19	11.05
水体	0.86	0.01	34.00	0.40
荒漠	12.81	0.34	3.09	326.52

地物类型	面积变化/km²			单一土地利用动态度/%
地物类型	1998—2008年	2008—2020年	1998—2020年	1998—2008年	2008—2020年	1998—2020年
农用地	+162.43	+22.77	+185.20	+1.96	+0.23	+1.01
建筑用地	+4.65	+32.12	+36.77	+2.00	+11.49	+7.17
水体	-5.49	+2.49	-3.00	-1.43	+0.76	-0.36
荒漠	-160.86	-56.83	-217.69	-2.87	-1.42	-1.76

[1]	罗镕基, 王宏涛, 王成. 基于改进遥感生态指数的甘肃省古浪县生态质量评价[J]. 干旱区地理, 2023, 46(4): 539-549.
[2]	卢文路, 刘志锋, 何春阳, 夏沛. 基于Sentinel-1A合成孔径雷达数据和全卷积网络的城市建设用地监测方法研究 [J]. 干旱区地理, 2020, 43(3): 750-760.
[3]	许亚平，王心源，王长林. 基于Landsat长时间序列数据的敦煌绿洲动态变化分析[J]. 干旱区地理, 2013, 36(5): 938-945.