基于无人机遥感影像和面向对象技术的荒漠草原植被分类

Vegetation classification of UAV remote sensing images in desert steppe based on object-oriented technology

探究适合荒漠草原植被遥感分类方法,明确荒漠草原地区植物物种类型及其分布状况,可以提高荒漠草原精细化生物多样性监测能力,对于荒漠草原的保护管理与生态可持续发展均具有重要意义。以贺兰山东麓洪积扇荒漠草原典型植被短花针茅、松叶猪毛菜、刺旋花、斑子麻黄为研究对象,利用无人机遥感影像,采用面向对象的分类回归树(classification and regression tree,CART)、K最邻近(K-nearest neighbor,KNN)、随机森林(random forest,RF)和支持向量机(support vector machine,SVM)分类方法,结合特征优选算法对影像特征进行优选,在此基础上选择最优特征进行荒漠草原植被精细化分类研究。结果表明:1)特征优选能够有效提高分类精度,应予以充分利用,当选取的特征组合为贡献度大于1.00%时,分类精度最高;2)基于无人机遥感影像挖掘的植被光谱、纹理特征,结合面向对象分类方法能有效实现贺兰山东麓荒漠草原典型植被精细化分类,其中RF分类精度最高,分类总体精度达到87.77%,Kappa系数为0.79。研究结果可为荒漠草原植被分类研究提供参考,对荒漠草原生物多样性保护管理与生态可持续发展均具有重要意义。

Exploring suitable remote sensing classification methods for desert grassland vegetation and clarifying the types of plant species and their distribution in desert grassland areas can improve the ability of fine biodiversity monitoring in desert grassland,which is of great significance for the protection and management of desert grassland as well as for the sustainable development of ecosystems within the landscape.This research studied the typical vegetation of Stipa breviflora,Convolvulus tragacanthoides,Salsola laricifolia,Ephedra rhytidosperma in the desert grassland of the floodplain fan at the eastern foothill of Helan Mountains.We used remote sensing images from unmanned aerial vehicles(UAVs),processed by object-oriented classification and regression tree(CART),Knearest neighbor(KNN),random forest(RF)and support vector machine(SVM)classification methods,combined with the feature selection algorithm to optimize the image features.We then selected the optimal features for the study of desert grassland vegetation classification,and its refinement.It was found that:1)Feature selection can effectively improve the classification accuracy and should be fully utilized,and the highest classification accuracy is achieved when the selected feature combination has a contribution degree greater than 1.00%;2)Based on the vegetation spectral and texture features mined from UAV remote sensing images,combining with the object-oriented classification method can effectively achieve the refined classification of the typical vegetation of the desert grassland at the eastern foothills of the Helan Mountains.In this research the RF classification accuracy was the highest,and the overall classification accuracy reached 87.77% with a Kappa coefficient of 0.79.The results of this study provide a reference for the study of vegetation classification in desert grasslands.This research will be of great significance for the conservation and management of desert grassland biodiversity and ecological sustainable development.