基于多源特征的工程廊道植被精细分类OA

To address the high-precision and high-timeliness requirements of ecological impact assessment for major engineering projects on vegetation cover,existing global and national land cover datasets exhibit insufficient adaptability in terms of spatial scale,classification system,and temporal alignment.Taking the Yangtze River-Huaihe River Water Diversion Project area as an example,leveraging Google Earth Engine,a multi-source feature set integrating spectral bands,spectral indices,GLCM textures,topographic factors,intra-annual quantiles,and IQR was constructed.A random forest algorithm was employed,and a high-pre-cision classification workflow was formed through feature importance-driven feature selection.Training and independent validation were completed based on Sentinel-2 10 m imagery and manually interpreted samples.The results demonstrate that the classification method integrating multi-source features and optimized selec-tion achieved high accuracy in the 2024 classification results,with an overall accuracy(OA)of 85.39%and a Kappa coefficient of 0.81,significantly outperforming the contemporaneous 10 m resolution Dynamic World product(OA 75.54%,Kappa 0.69),particularly in capturing details of complex scenarios such as engineering corridors and narrow shoreline areas.Feature importance analysis revealed that intra-annual quantiles and IQR phenological features,water and soil-enhanced indices,SWIR bands,and texture indica-tors contributed prominently to accuracy improvement.In 2024,the main project line had formed a continu-ous water body belt,with hardened embankments and dams along the route creating narrow impervious zones,and localized conversions of farmland and woodland to bare land or sparse vegetation,which warrant special attention.In conclusion,this method exhibits significant advantages in classification accuracy,time-liness,and engineering detail recognition,providing high-quality baseline support for ecological impact as-sessment of major engineering projects.