森林是陆地生态系统的主体，我国林地资源分布复杂，传统森林变化检测存在工作量大、效率低下等诸多问题，难以满足新时期林地变化检测任务的需求。随着对地观测技术的发展及应用，将多源遥感数据与变化检测技术结合，快速准确获取森林变化信息，对森林资源管理和保护利用具有重要现实意义。以甘肃省文县局部区域为研究区，利用Sentinel-1、Sentinel-2、DEM遥感数据，结合地面样地资料，构建机器学习分类算法模型，研究多源数据在地物分类中的应用潜力；对特征变量分析优化获得研究区分类精度最高的林地提取结果；研究基于不同算法、数据源和平台下的林地变化检测技术并分析各方法特点与适用性。主要研究内容和结论如下：

（1）应用GEE平台与CollectEarth“增强目视解译法”高效率完成研究区数据获取，预处理及样本点布设核查工作；提取Sentinel-1、Sentinel-2、DEM多源遥感数据的光谱、指数、纹理、后向散射系数、地形特征；应用Jeffries-Mausita（J-M距离）对两期影像构建的72个特征变量择优筛选，按照特征类型设计6种组合方案，实验并分析各特征组合对地物分类精度的影响。研究表明，相对于仅使用单一数据源分类，雷达影像与光学影像的结合可以有效提升地物分类的效果。

（2）以特征重要性为指标，对分类精度最高的特征组合排序，逐特征变量组合分类，在GEE平台的支持下，快速发掘出分类精度最高的特征变量组合，实现双时相研究区影像林地精细化提取工作；对比基于GEE平台多源遥感随机森林法与仅使用Sentinel-2原始影像的林地变化提取结果。研究表明，基于多源遥感随机森林法变化检测精度更高。

（3）研究不同林地变化检测方法适用性，指数特征法具有原理简单、速度快等优势，但存在阈值设定困难等问题；深度学习法具有自动化程度高、受人为因素影响小等优势，但存在模型训练时间长，精度易受样本影响等问题；基于GEE平台多源遥感随机森林法林地变化检测，具有样本采集速度快、分类模型构建效率高、计算成本低廉等优势，符合当前林地变化检测应将遥感、大数据、模型算法与创新技术结合实现的总体目标。

Forest is the main body of terrestrial ecosystem. The distribution of forest resources in China is complex. The traditional forest change detection has many problems such as heavy workload and low efficiency, which is difficult to meet the needs of forest land change detection tasks in the new period. With the development and application of earth observation technology, combining multi-source remote sensing data with change detection technology to quickly and accurately obtain forest change information has important practical meaning for forest resource management and protection and utilization. This paper takes the local area of Wenxian County in Gansu Province as the research area, uses Sentinel-1, Sentinel-2 and DEM remote sensing data, and combines the ground sample data to construct the machine learning classification algorithm model, and studies the application potential of multi-source data in ground feature classification. The extraction results of woodland with the highest classification accuracy in the study area were obtained by analyzing and optimizing the characteristic variables. Research on forest land change detection technology based on different algorithms, data sources and platforms, and analyze the characteristics and applicability of each method. The main research contents and conclusions of this paper are as follows:

(1)Using GEE platform and CollectEarth of enhanced visual interpretation method to complete the data acquisition, preprocessing and sample point layout verification in the study area efficiently; extraction of Sentinel-1, Sentinel-2, DEM multi-source remote sensing data spectrum, index, texture, backscatter coefficient, terrain features; Jeffries-Mausita(J-M distance) was used to select 72 feature variables from two images, and six combination schemes were designed according to the feature types to study and analyse the influence of each feature combination on the classification accuracy of ground objects. Studies have shown that the combination of radar images and optical images can effectively improve the effect of ground object classification compared with the single data source classification.

(2)Taking the feature importance as the index, sorting the feature combination with the highest classification accuracy, combining the classification by feature variables, and quickly excavating the feature variable combination with the highest classification accuracy under the support of GEE platform, so as to realize the fine extraction of image woodland in the two-phase study area; the results of multi-source remote sensing random forest method based on GEE platform and Sentinel-2 original spectral forest land change extraction were compared. The research shows that the change detection accuracy of random forest method based on multi-source remote sensing is higher.

(3)The applicability analysis of different forest land change detection methods is studied. Exponential feature method has the advantages of simple algorithm principle and fast calculation speed, but it is difficult to set threshold. Deep learning method has the advantages of high degree of automation of model training and small influence of human factors on algorithm. However, there are some problems such as long training time and easily affected by samples. In this paper, the forest land change detection based on multi-source remote sensing random forest algorithm on GEE platform has the advantages of fast sample collection speed, high classification model construction efficiency and low calculation cost. It is in line with the overall goal of the current forest land change detection that remote sensing, big data, model algorithm and innovative technology should be combined to achieve.

[1]吴霜, 延晓冬, 张丽娟. 中国森林生态系统能值与服务功能价值的关系[J]. 地理学报, 2014, 69(3): 334-342.

[2]Beer C, Reichstein M, Tomelleri E, et al. Terrestrial gross carbon dioxide uptake: global distribution and covariation with climate[J]. Science, 2010, 329(5993): 834-838.

[3]Weismiller R A. Change detection in coastal zone environments[J]. Photogrammetric Engineering and Remote Sensing, 1977, 43(12): 1533-1539.

[4]Royle D D, Lathrop R G. Discriminating Tsuga canadensis hemlock forest defoliation using remotely sensed change detection[J]. Journal of Nematology, 2002, 34(3): 213-221.

[5]Dianat R, Kasaei S. Change detection in optical remote sensing images using difference-based methods and spatial information[J]. IEEE Geoscience and Remote Sensing Letters, 2009, 7(1): 215-219.

[6]Nackaerts K, Vaesen K, Muys B, et al. Comparative performance of a modified change vector analysis in forest change detection[J]. International Journal of Remote Sensing, 2005, 26(5):839-852.

[7]Marinelli D, Coops N C, Bolton D K, et al. Forest Change Detection in Lidar Data Based on Polar Change Vector Analysis[J]. IEEE Geoscience and Remote Sensing Letters, 2022, 19(99):1-5.

[8]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.

[9]Chen C F, Nguyen-Thanh S, Chang N B, et al. Multi-Decadal Mangrove Forest Change Detection and Prediction in Honduras, Central America, with Landsat Imagery and a Markov Chain Model[J]. Remote Sensing, 2013, 5(12):6408-6426.

[10]Munyati C. Wetland change detection on the Kafue Flats, Zambia, by classification of a multitemporal remote sensing image dataset[J]. International journal of remote sensing, 2000, 21(9): 1787-1806.

[11]罗號, 孙华, 胡满,等. 基于Landsat影像的黄丰桥林场森林变化检测研究[J]. 中南林业科技大学学报, 2017, 37(12): 65-71.

[12]Descl E E B, Bogaert P, Defourny P. Forest change detection by statistical object-based method[J]. Remote sensing of Environment, 2006, 102(1): 1-11.

[13]李春干, 梁文海. 基于面向对象变化向量分析法的遥感影像森林变化检测[J]. 国土资源遥感, 2017, 29(3): 77-84.

[14]Wang F, Xu Y J. Comparison of remote sensing change detection techniques for assessing hurricane damage to forests[J]. Environmental Monitoring and Assessment, 2010, 162(1):311-326.

[15]许石罗, 牛瑞卿, 武雪玲, 等. 基于主成分分析与粒子群优化的遥感影像变化检测[J]. 测绘科学, 2017, 42(4): 151-156.

[16]许璐. 中国东北部边境森林植被覆盖度变化研究[D]. 青海: 青海师范大学, 2018.

[17]王崇阳, 田昕. 基于GF-1PMS数据的森林覆盖变化检测[J]. 遥感技术与应用, 2021, 36(1): 208-216.

[18]张劳模. 基于深度学习的林地分类与动态变化模拟技术研究[D]. 北京: 中国林业科学研究院, 2020.

[19]Khan, SH, XM, et al. Forest Change Detection in Incomplete Satellite Images With Deep Neural Networks[J]. IEEE Transactions on Geoscience and Remote Sensing, 2017,55(9):5407-5423.

[20]卢泽如, 熊勤学, 周雨顺, 等. 运用Sentinel-1A时序数据和动态时间规整算法提取稻虾田空间分布信息[J]. 江苏农业科学, 2020, 48(18): 230-236.

[21]刘睿, 王志勇, 高瑞. 时序SAR影像的干旱地区土地利用分类应用[J]. 测绘科学, 2021, 46(10): 90-97.

[22]栾玉洁, 郭金运, 高永刚, 等. 基于Sentinel-1B SAR数据的2018年寿光洪水遥感监测及灾害分析[J]. 自然灾害学报, 2021, 30(2): 168-175.

[23]韩可欣, 刘绥华, 许璟, 等. Sentinel-1A数据在高原山区短期土地利用监测中的应用——以贵州省安顺市为例[J]. 西南大学学报(自然科学版), 2017, 39(4): 124-130.

[24]Ghasemi M, Karimzadeh S, Feizizadeh B. Urban classification using preserved information of high dimensional textural features of Sentinel-1 images in Tabriz, Iran[J]. Earth Science Informatics, 2021, 14(4): 1745-1762.

[25]Beriaux E, Jago A, Lucau-Danila C, et al. Sentinel-1 Time Series for Crop Identification in the Framework of the Future CAP Monitoring[J]. Remote Sensing, 2021, 13(14): 2785-2814.

[26]李晓慧, 王宏, 李晓兵, 等. 基于多时相Landsat 8 OLI影像的农作物遥感分类研究[J]. 遥感技术与应用, 2019, 34(2): 389-397.

[27]刘志勇, 黄宇飞, 刘佳欣, 等. 应用Landsat-8卫星遥感图像的青海地区土壤分类研究[J]. 航天器工程, 2020, 29(1): 105-109.

[28]黄鹏程, 张明明, 王新宇, 等. 基于Landsat-8 OLI的西安市土地利用类型遥感分类研究[J]. 测绘与空间地理信息, 2020, 43(1): 85-88.

[29]刘瑞清, 李加林, 孙超, 等. 基于Sentinel-2遥感时间序列植被物候特征的盐城滨海湿地植被分类[J]. 地理学报, 2021, 76(7): 1680-1692.

[30]邢瑾, 候建西, 刘勇, 等. 基于Sentinel-2数据的祁连山草地自动提取策略[J]. 兰州大学学报(自然科学版), 2021, 57(4): 473-482.

[31]张东彦, 戴震, 徐新刚, 等. 基于时序Sentinel-2影像的现代农业园区作物分类研究[J]. 红外与激光工程, 2021, 50(5): 262-272.

[32]Brinkhoff J, Vardanega J, Robson A J. Land cover classification of nine perennial crops using sentinel-1 and-2 data[J]. Remote Sensing, 2020, 12(1): 96-122.

[33]Steinhausen M J, Wagner P D, Narasimhan B, et al. Combining Sentinel-1 and Sentinel-2 data for improved land use and land cover mapping of monsoon regions[J]. International journal of applied earth observation and geoinformation, 2018, 73(1): 595-604.

[34]Mercier A, Betbeder J, Rumiano F, et al. Evaluation of Sentinel-1 and 2 time series for land cover classification of forest-agriculture mosaics in temperate and tropical landscapes[J]. Remote Sensing, 2019, 11(8): 979-999.

[35]智超, 吴文挺, 苏华. 潮汐和植被物候影响下的潮间带湿地遥感提取[J]. 遥感学报, 2022, 26(2): 373-385.

[36]熊皓丽, 周小成, 汪小钦, 等. 基于GEE云平台的福建省10 m分辨率茶园专题空间分布制图[J]. 地球信息科学学报, 2021, 23(7): 1325-1337.

[37]宁晓刚, 常文涛, 王浩, 等. 联合GEE与多源遥感数据的黑龙江流域沼泽湿地信息提取[J]. 遥感学报, 2022, 26(2): 386-396.

[38]付甜梦, 张丽, 陈博伟, 等. 基于GEE平台的海岛地表覆盖提取及变化监测——以苏拉威西岛为例[J]. 遥感技术与应用, 2021, 36(1): 55-64.

[39]Stromann O, Nascetti A, Yousif O, et al. Dimensionality reduction and feature selection for object-based land cover classification based on Sentinel-1 and Sentinel-2 time series using Google Earth Engine[J]. Remote Sensing, 2020, 12(1): 76-100.

[40]王志伟, 杨国东, 张旭晴, 等. 高分六号卫星遥感影像不同几何校正方法精度对比研究[J]. 世界地质, 2021, 40(1): 125-130.

[41]高雪艳, 潘安宁, 杨扬. 基于图像混合特征的城市绿地遥感图像配准[J]. 浙江大学学报(工学版), 2019, 53(6): 1205-1217.

[42]杨晓涵, 李雨彤, 王宁, 等. 遥感影像融合方法比较研究[J]. 重庆工商大学学报(自然科学版), 2017, 34(2): 90-94.

[43]唐红梅, 李永川. 基于主成分变换的遥感影像变化检测[J]. 四川建材, 2014, 40(6): 230-231.

[44]张路. 基于多元统计分析的遥感影像变化检测方法研究[D]. 武汉: 武汉大学, 2004.

[45]何云, 黄翀, 李贺, 等. 基于Sentinel-2A影像特征优选的随机森林土地覆盖分类[J]. 资源科学, 2019, 41(5): 992-1001.

[46]刘贾贾, 刘志辉, 李凤. 基于遥感影像的城镇建筑物群分类[J]. 自然灾害学报, 2021, 30(6): 61-66.

[47]杨雪峰, 吐热尼古丽·阿木提. 基于人工蜂群算法优化的SVM遥感分类方法——以玛纳斯湖古湖盆为例[J]. 地理与地理信息科学, 2018, 34(4): 40-45.

[48]李楠, 朱秀芳, 潘耀忠, 等. 人工蜂群算法优化的SVM遥感影像分类[J]. 遥感学报, 2018, 22(4): 559-569.

[49]虞瑶, 苏红军, 姚文静. 基于Boosting的高光谱遥感切空间协同表示集成学习方法[J]. 遥感技术与应用, 2020, 35(3): 634-644.

[50]Bao R, Xia J, Dalla Mura M, et al. Combining morphological attribute profiles via an ensemble method for hyperspectral image classification[J]. IEEE Geoscience and Remote Sensing Letters, 2016, 13(3): 359-363.

[51]侯蒙京, 殷建鹏, 葛静, 等. 基于随机森林的高寒湿地地区土地覆盖遥感分类方法[J]. 农业机械学报, 2020, 51(7): 220-227.

[52]蒲东川, 王桂周, 张兆明,等. 基于独立成分分析和随机森林算法的城镇用地提取研究[J]. 地球信息科学学报, 2020, 22(8): 1597-1606.

[53]Belgiu M, Dr U A Gu C T L. Random forest in remote sensing: A review of applications and future directions[J]. ISPRS journal of photogrammetry and remote sensing, 2016, 100(114): 24-31.

[54]宋荣杰, 宁纪锋, 常庆瑞, 等. 基于小波纹理和随机森林的猕猴桃果园遥感提取[J]. 农业机械学报, 2018, 49(4): 222-231.

[55]沈江龙, 郑江华, 尼格拉·吐尔逊, 等. 若羌绿洲特色林果种植信息遥感提取方法适用性分析[J]. 中国农业资源与区划, 2022, 43(2): 206-219.

[56]Rojas R. Neural networks: a systematic introduction[M]. Berlin: Springer Science & Business Media, 2013: 3-5.

[57]刘树栋, 魏嘉敏. 基于谱聚类和成对数据表示的多层感知机分类算法[J]. 计算机科学, 2019, 46(2): 194-198.

[58]杜昕, 范廷恩, 董建华, 等. 基于多层感知机网络的薄储层预测[J]. 石油地球物理勘探, 2020, 55(6): 1178-1187.

[59]张刚. 基于深度学习的遥感图像语义分割关键技术研究[D]. 四川: 中国科学院大学(中国科学院光电技术研究所), 2020.

[60]张珂, 冯晓晗, 郭玉荣, 等. 图像分类的深度卷积神经网络模型综述[J]. 中国图象图形学报, 2021, 26(10): 2305-2325.

[61]刘修宇, 江涛, 厉彦一. 基于机器学习中分辨率遥感影像分类应用研究[J]. 地理信息世界, 2021, 28(4): 66-73.

[62]张功国, 吴建, 易亿, 等. 基于集成卷积神经网络的交通标志识别[J]. 重庆邮电大学学报: 自然科学版, 2019, 31(4): 571-577.

[63]沈锐, 陈亚军. 一种基于批量归一化的LeNet网络改进方法[J]. 四川文理学院学报, 2021, 31(2): 136-140.

[64]郭颖, 李增元, 陈尔学, 等. 一种改进的高空间分辨率遥感影像森林类型深度学习精细分类方法:双支FCN-8s[J]. 林业科学, 2020, 56(3): 48-60.

[65]唐少飞, 田庆久, 徐凯健, 等. Sentinel-2卫星落叶松林龄信息反演[J]. 遥感学报, 2020, 24(12): 1511-1524.

[66]张宏建, 王冰, 周健, 等. 基于BP神经网络的内陆河流水质遥感反演[J]. 华中师范大学学报(自然科学版), 2022, 56(2): 333-341.

[67]Zhang Z, Liu Q, Wang Y. Road extraction by deep residual u-net[J]. IEEE Geoscience and Remote Sensing Letters, 2018, 15(5): 749-753.

[68]Shamsolmoali P, Zareapoor M, Wang R, et al. A novel deep structure U-Net for sea-land segmentation in remote sensing images[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2019, 12(9): 3219-3232.

[69]向阳, 赵银娣, 董霁红. 基于改进UNet孪生网络的遥感影像矿区变化检测[J]. 煤炭学报, 2019, 44(12): 3773-3780.

[70]赵亚红, 江培华, 李佳慧,等. 基于Sentinel-2影像的SAVI快速过火区面积提取[J]. 华北科技学院学报, 2021, 18(6): 27-34.

[71]Royle D D, Lathrop R G. Discriminating Tsuga canadensis hemlock forest defoliation using remotely sensed change detection[J]. Journal of Nematology, 2002, 34(3): 213-215.

[72]Louhaichi M, Borman M M, Johnson D E. Spatially located platform and aerial photography for documentation of grazing impacts on wheat[J]. Geocarto International, 2001, 16(1): 65-70.