森林工程2025，Vol.41Issue(6)：1101-1115,15.DOI:10.7525/j.issn.1006-8023.2025.06.001
基于Sentinel-2多维特征融合的铁岭市优势树种分类
Classification of Dominant Tree Species in Tieling Based on Sentinel-2 Multi-Dimensional Feature Fusion
摘要
Abstract
To address the challenge of Sentinel-2 data in distinguishing spectrally similar tree species,this study estab-lished a multi-dimensional feature fusion classification system based on the google earth engine(GEE)cloud platform,with Tieling City,Liaoning Province as the experimental area.By integrating Sentinel-2 multi-temporal data and employ-ing multi-dimensional feature statistical methods,we extracted spectral and vegetation index features including quan-tiles,extremes,and standard deviations,combined with topographic,textural,phenological,and harmonic features,forming a total of 120 features across six categories.Multiple feature combination schemes were designed and imple-mented through a hierarchical classification strategy using the random forest algorithm,ultimately achieving fine classifi-cation of seven dominant tree species:Pinus tabuliformis,Pinus sylvestris var.mongolica,Larix gmelinii,Populus,fruit trees,Quercus mongolica,and Robinia pseudoacacia.The results demonstrated that multi-dimensional temporal statisti-cal features effectively captured subtle interspecies differences.Variations in water content between Pinus sylvestris var.mongolica and Pinus tabuliformis were successfully characterized through multiple vegetation indices.Topographic and textural features played decisive roles in distinguishing deciduous species.The classification overall accuracy reached 94.7%for evergreen species and 88.1%for deciduous species,with all six feature combination schemes achieving over-all accuracy exceeding 77.9%.This study confirms that the integration of multi-dimensional feature statistical methods with the GEE platform fully exploits the multi-band advantages of Sentinel-2 data,significantly enhancing large-scale tree species classification capabilities through temporal feature analysis.It provides a cost-effective solution for dynamic monitoring of forest resources at large-scale,with the cloud-based processing framework demonstrating potential for appli-cation expansion to broader geographical regions.关键词
树种分类/时序统计特征/多特征融合/随机森林算法/谷歌地球引擎(google earth engine,GEE)Key words
Tree species classification/temporal statistical features/multi-feature fusion/random forest/google earth engine(GEE)分类
林学引用本文复制引用
刘晟屹,温一博,武锦炜,常文龙,彭代亮..基于Sentinel-2多维特征融合的铁岭市优势树种分类[J].森林工程,2025,41(6):1101-1115,15.基金项目
国家自然科学基金(32101518) （32101518）
辽宁省林业和草原局科技创新平台研发项目(LLC20224). （LLC20224）
