College of Forestry,Northeast Forestry University,Harbin 150040,China
Show less
文章历史+
Received
Published
2024-12-04
2025-09-15
Issue Date
2025-10-30
PDF (2436K)
摘要
准确掌握森林覆盖空间分布对于森林生态系统保护、恢复和可持续利用至关重要。但高效、精准地获取县域尺度复杂森林覆盖变化依靠低空间分辨率遥感影像结合传统计算机分类模型已经无法满足。以黑龙江省佳木斯汤原县复杂森林为研究对象,采用哨兵一号、二号(Sentinel-1、Sentinel-2)中空间分辨率卫星遥感影像,构建基于粒子群优化算法(particle swarm optimization,PSO)优化的机器学习模型,检测县域尺度森林覆盖变化,应用K-折交叉验证对检测森林覆盖结果进行精度评价。研究结果表明,基于粒子群算法优化的支持向量机和随机森林2个机器学习模型与未经参数优化的自身模型相比,森林覆盖变化检测精度均得到提高,支持向量机模型提高6.52%,随机森林模型提高4.65%。与目前主流ESA World Cover土地覆盖产品相比,基于粒子群算法优化的随机森林模型精度最高,总体精度达到0.92。优化后的随机森林模型对森林覆盖变化检测也更加精细。通过粒子群优化算法的随机森林模型对中空间分辨率遥感影像进行分类,可以快速、准确地掌握县域尺度森林覆盖空间分布情况,为森林生态系统保护、恢复和可持续利用提供数据和技术支撑。
Abstract
Accurately grasping the spatial distribution of forest cover is crucial for the protection, restoration and sustainable use of forest ecosystems. However, it is no longer possible to efficiently and accurately obtain the changes in complex forest cover at the county scale by relying on low spatial resolution remote sensing images combined with traditional computer classification models. Therefore, this study took the complex forests in Tangyuan County, Jiamusi, Heilongjiang Province as the research object, used the medium spatial resolution satellite remote sensing images of Sentinel-1 and Sentinel-2, and constructed a machine learning model optimized by particle swarm optimization (PSO) to detect the changes in forest cover at the county scale. The K-fold cross validation was used to evaluate the accuracy of the forest cover detection results. The results showed that the support vector machine and random forest machine learning models optimized by particle swarm algorithm had improved the accuracy of forest cover change detection compared with their own models without parameter optimization. The support vector machine model increased by 6.52%, and the random forest model increased by 4.65%. Compared with the current mainstream ESA COVER WORD land cover product, the random forest model optimized by particle swarm algorithm had the highest accuracy, with an overall accuracy of 0.92. The optimized random forest model was also more precise in detecting forest cover changes. By classifying medium spatial resolution remote sensing images through the random forest model of the particle swarm optimization algorithm, we can quickly and accurately grasp the spatial distribution of forest cover at the county scale, and provide data and technical support for the protection, restoration and sustainable utilization of forest ecosystems.
根据全国遥感检测土地利用覆盖分类体系,结合汤原县实际情况,将一级分类中的城乡、工矿、居民用地和未利用土地合成建筑用地,将林地分为森林和灌木林。因此按照类别将汤原县土地覆盖分为森林、灌木林、草地、农田、水域和建设用地6个类别。为了保证森林覆盖变化检测模型选择训练和验证样本的随机性,应用ArcGIS 10.4软件,设置样本点之间的间距不超过500 m,在汤原县内生成了2 371个随机点,样本的类别和数目见表1。样本点通过结合Google Earth上的高分影像、汤原县高分数据及地类图斑数据和专家知识来标注。为进一步比较基于粒子群优化算法(PSO)优化的机器学习模型对森林覆盖变化检测效果,选择现有主流ESA World Cover土地利用数据提取森林覆盖变化进行比较。ESA World Cover是欧洲航天局发布的产品,此产品是同样是基于哨兵Sentinel-1 C波段和Sentinel-2多光谱L2A地表反射率图像数据生成,与本研究检测的森林覆盖变化具有可比性和客观性。
本研究采用K-折交叉验证方法将2 371个样本点随机分为K份,依次使用其中的(K-1)份进行训练,使用剩下的1份作为测试集,本研究将K取值为5。K-折交叉验证可以有效地避免数据产生过拟合和欠拟合现象,可以有效地评估模型的质量[24]。森林覆盖变化检测精度评价采用指标为总体精度(OA,式中记为OA)和Kappa系数(式中记为Kappa)[25]。OA和Kappa系数是通过分类结果与验证样本、三调地类图斑数据、ESA World Cover土地利用数据对比,构建精度评价混淆矩阵获得。其公式为
采用基于粒子群算法的参数优化的和支持向量机和随机森林模型对2020年和2021年汤原县Sentinel-1和Sentinel-2遥感影像进行分类,分类精度见表4。优化后的支持向量机和随机森林模型总精度和Kappa在2个年份上均高于ESA产品精度,其中,2020和2021年随机森林模型总精度分别提高了11.27%和11.98%,2020和2021年支持向量机模型总精度分别提高了7.12%和6.78%。与现有森林覆盖产品的局部进行对比,研究分别取汤原西部森林集中区域、中部农田区域以及东部水域附近区域。对比结果如图4所示,在森林集中区域中SVM和RF模型更能识别出其中的其他地类。对于农田及道路周边的森林提取效果更明显,所提取的森林图斑更加完整准确。随机森林模型和支持向量机模型相比ESA World Cover而言,在处理水域附近湿地类型时,都存在出现将草地错分的现象。综上所述,随机森林模型分类结果图斑相对产品和支持向量机模型更加细致精准。
XIET Y, PANJ, SUNY L,et al.Classification of forest tree species using multi-dimensional remote sensing information coordination[J].Journal of Northeast Forestry University,2023,51(3):73-78.
[3]
ZHANGY, PENGC, LIW Z,et al.Monitoring and estimating drought-induced impacts on forest structure,growth,function,and ecosystem services using remote-sensing data:recent progress and future challenges [J].Environmental Reviews,2013(21):103-115.
WUX Q, QINX L, ZHOUR L,et al.Progress of study on forest cover change detection by using remote sensing technique[J].Forest and Grassland Resources Research,2010(4):82-87.
[6]
陈鑫镖.遥感影像变化检测技术发展综述[J].测绘与空间地理信息,2012,35(9):38-41.
[7]
CHENX B.A summary of change detection techniques of remote sensing imagery[J].Geomatics & Spatial Information Technology,2012,35(9):38-41.
[8]
HAOR X, CHENE X, LIZ Y.Forest cover change detection method using bi-temporal GF-1 multi-spectral data[C]//2016 IEEE International Geoscience and Remote Sensing Symposium(IGARSS),10-15 July 2016,Beijing,China.IEEE,2016:7729975.
RENC, JUH B, ZHANGH Q,et al.Research on remote sensing monitoring technology of forest land dynamic change in Tianshui in recent 30 Years[J].Forest Research,2017,30(1):25-33.
JIAX F, ZHUH C, LINGF,et al.Forest cover monitoring and its changes in Hanjiang River basin based on Landsat multispectral and PALSAR/PALSAR-2 data[J].Resources and Environment in the Yangtze Basin,2021,30(2):321-329.
XIONGX Y, YANGX Z, ZHAOY C,et al.Remote sensing inversion of forest biomass based on hyperparametric optimized random forests algorithm[J].Journal of Central South University of Forestry & Technology,2024,44(5):102-111.
[19]
THIENB B, PHUONGV T.Using landsat satellite imagery for assessment and monitoring of long-term forest cover changes in Dak Nong Province,Vietnam[J].Geographica Pannonica,2023,27(1):69-82.
[20]
HECKELK, URBANM, SCHRATZP,et al.Predicting forest cover in distinct ecosystems:The potential of multi-source sentinel-1 and -2 data fusion[J].Remote Sensing,2020,12(2):302.
YANG D, ZUOX M, CHENJ,et al.Classification of forest stand types in Sentinel-2 remote sensing images based on multi-feature optimization[J].Forest Engineering,2023,39(3):12-20.
[23]
FIORAVANTEP D, LUTIT, CAVALLIA,et al.Multispectral Sentinel-2 and SAR Sentinel-1 integration for automatic land cover classification[J].Land,2021,10(6):611.
[24]
MERCIERA, BETBEDERJ, RUMIANOF,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.
ZHAOL W, FANW Y, NIEY H.Estimating forest biomass using dual-frequency synthetic aperture radar (SAR) data and an interferometric water cloud model[J].Journal of Northeast Forestry University,2024,52(9):58-68.
[27]
HADJIMITSISD G, THEMISTOCLEOUSK, MICHAELIDESS,et al.Tasselled cap transform for change detection in the drylands:findings for spot and landsat satellites with foss tools[J].International Society for Optics and Photonics,2015,9535:953512.
ZHAOB X, ZHANGY.Study on vegetation information extraction method based on Landsat-8 OLI images[J].Geomatics & Spatial Information Technology,2018,41(1):79-82,85.
MUZ Y, ZHANGZ P, ZHANGH,et al.Applying machine learning algorithm models to predict aboveground biomass of Larix gmelinii in Xing'an[J].Journal of Northeast Forestry University,2024,52(3):41-47.
DUANZ L, ZHOUX.Research on tea resources in Zhenyuan county based on remote sensing technology[J].Journal of West China Forestry Science,2024,53(5):140-145.
ZHANGX Y, LIF R, ZHENZ,et al.Forest vegetation classification of Landsat-8 remote sensing image based on random forests model[J].Journal of Northeast Forestry University,2016,44(6):52-57,74.
[36]
LACHAUDA, ADAMM, MIŠKOVIĆI.Comparative study of random forest and support vector machine algorithms in mineral prospectivity mapping with limited training data[J].Minerals,2023,13(8):1073.
LIUX G, XIAOX, LANY B,et al.Forest vegetation segmentation method with UAV visible light remote sensing images[J].Journal of Northeast Forestry University,2023,51(4):62-67.
[39]
TIANY, ZHANGT, XIAOJ H,et al.A coevolutionary framework for constrained multiobjective optimization problems[J].IEEE Transactions on Evolutionary Computation,2021,25(1):102-116.
[40]
BERGMEIRC, HYNDMANR J,KOO B.A note on the validity of cross-validation for evaluating autoregressive time series prediction[J].Computational Statistics & Data Analysis,2017,120:70-83.
[41]
GUOY T, JIAOW L, LONGT F,et al.Generation of forest cover map of the Qinghai-Tibet Plateau based on multisource dataset and random forest[J].Frontiers in Environmental Science,2022,10:939151.
FANY L, TANGS N, TANB X.Forest cover change detection based on multi-scale segmentation and tasseled cap transformation over plateau area[J].Journal of Beijing Forestry University,2023,45(4):60-69.
[44]
CAFFARATTIG D, MARCHETTAM G, EUILLADESL D,et al.Improving forest detection with machine learning in remote sensing data[J].Remote Sensing Applications:Society and Environment,2021,24:100654.
[45]
ADUGNAT, XUW, FANJ.Comparison of random forest and support vector machine classifiers for regional land cover mapping using coarse resolution FY-3C images[J].Remote Sensing,2022,14(3):574.
WANGJ, LIH K, LONGB P,et al.Tree species identification in southern hilly areas using sentinel series images with multi-temporal features[J].Journal of Northeast Forestry University,2024,52(3):60-68.