基于GEE的多源遥感影像特征优选方法的水稻种植区信息提取

打开文本图片集

中图分类号：TP79;S127 文献标识码：A 文章编号：1000-4440（2025）06-1159-10

Abstract：Multi-temporal Sentinel-1/2datademonstratesignificantadvantagesinidentifyingandmonitoringrice cultivation incomplex mountainousandhillyareas，aswellasincloudyandrainyenvironments，which providerich information forriceidentification.However，anexcesivenumberoffeature variablesmayleadtodimensionaldisastersandinformationredundancy.Inthis study，a featureselectionmethod wasemployedtoevaluate therecognitionaccuracyoffive feature combination schemes（spectral features，spectral features + vegetation indices，spectral features + vegetation indices + texture features，spectral features + vegetation indices + texture features + radarinformation，and feature selection） forricecultivationarea，using Sentinel-1/2multi-spectralandmulti-temporaldata.Thespatial mappingacuracyof each schemewas alsoanalyzed.Theresults indicatedthat Scheme five，which incorporatedfeature selection，performed the best in rice identification，with an overall accuracy of 92.60% ，a Kappa coefficient of O.903 O，and an F1 score of 92.40 % ： Comparedwiththericecultivationareadatain2O24fromtheJiangxi Statistical Yearbook，Schemefiveachievedahighaccuracy of 98.73% in estimating the rice area in Jiangxi province，with a significantly lower relative error compared to

Schemeone-Scheme four.Theresults of this studyconfirm that the integrated application ofmulti-source remote sensing data and multi-temporal feature selection methods caneffectively reduce data redundancy and enhance the accuracy and precision of rice area extraction.

Key words：multi-source remote sensing；rice； Jiangxi province；feature selection；random forest method

粮食安全是国家安全与社会稳定的基石，党的二十大报告强调“全方位夯实粮食安全根基”的重要性，对保障粮食安全提出更高的要求[1]。（剩余12846字）