基于高光谱MNF-SVM法荒漠化草原地表微斑块识别研究

Recognition of Surface Micro-Patches on Desertified Grassland Based on Hyperspectral MNF-SVM Method

植被群落结构和土壤性质改变是草原退化的主要表现,鼠害是加剧草原退化的重要因素,对地表微斑块识别是草原退化研究的基础。本文利用GaiaiSky-mimi型高光谱仪采集典型荒漠化草原高光谱影像,使用主成分分析法(Principal Components Analysis,PCA)与最小噪声分离法(Minimum Noise Fraction,MNF)对数据进行前期处理,后采用支持向量机(Support Vector Machine,SVM)法进行地表微斑块识别,并与迭代自组织数据分析法(Isodata)和K均值聚类法(K-means)识别结果进行比较。结果表明:经过MNF处理后,支持向量机法的总体识别精度和Kappa系数最高,识别结果优于其它2种识别方法,对目前处理荒漠化草原地表微斑块的识别具有准确性和实用性,同时为无人机高光谱遥感提供数据及理论支持。

The changes in vegetation community structure and soil properties are the main manifestations of the grassland degradation,and the rodent damage is an important factor aggravating the grassland degradation. In this paper,the GaiaiSky-mimi high spectrometer was used to collect the hyperspectral images of typical desertified grasslands,and the principal components analysis(PCA)and minimum noise fraction(MNF)methods were used to preliminary process the data,then the support vector machine(SVM)method was used to identify the surface micro-patches. The results were compared with those obtained by the iterative self-organizing data analysis and K-means clustering methods. The results showed that the overall recognition accuracy and Kappa coefficient of the support vector machine method were the highest after the MNF treatment,and the recognition results were superior to those of the other two methods. It was accurate and practical for the current recognition of surface micro-patches on the desertified grasslands,providing data and theoretical supports for the UAV hyperspectral remote sensing.