“珠海一号”高光谱遥感数据在土地利用分类中的潜力研究

Potential Study of Zhuhai-1 Hyperspectral Remote Sensing Data in Land Use Classification

探究高光谱遥感数据在土地利用分类中的作用,有助于全面了解土地资源的现状和动态变化,为环境保护、城市规划和农业管理提供基础数据。以国产“珠海一号”高光谱遥感数据为主,使用最小噪声变换处理后,对比分析了最大似然法、支持向量机、随机森林和面向对象4种分类方法的地物分类精度。结果表明:分类的总体精度由高到低依次为随机森林(76.77%)、最大似然法(72.32%)、支持向量机(67.25%)和面向对象(49.20%),对应的Kappa系数分别为0.70、0.65、0.59、0.39。随后,采用分类精度较高的随机森林方法对验证区域进行分类,并得出验证区域的总体精度为72.03%,Kappa系数为0.61。随机森林方法在高光谱遥感影像土地利用分类中表现出较高的分类精度,可为未来土地利用分类的研究提供参考。

Exploring the role of hyperspectral remote sensing data in land use classification can help to comprehensively understand the current status and dynamic changes of land resources,provide accurate basic data for environmental protection,urban planning and agricultural management.Using primarily the domestic Zhuhai-1 hyperspectral remote sensing data and after applying the minimum noise fraction transformation,we compared and analyzed the classification accuracy of four classification methods:maximum likelihood(ML),support vector machine(SVM),random forest(RF)and object-oriented classification(OBC).The results indicated that the overall classification accuracy from highest to lowest is RF(76.77%),ML(72.32%),SVM(67.25%),OBC(49.20%).The corresponding Kappa coefficients were 0.70,0.65,0.59,and 0.39,respectively.Further application of the RF algorithm,which demonstrated the highest classification accuracy,resulted in an overall accuracy of 72.03%and a Kappa coefficient of 0.61.This indicated that the RF method ensures high classification accuracy in hyperspectral remote sensing imagery and provides a reference for land use classification.