以艾比湖流域为研究区域,典型盐渍土为研究对象,引入分数阶微分,以0.2为微分阶数间隔,将0~2细分为11阶微分,对原始光谱反射率及其常用的均方根、倒数等数学变换进行微分计算,结合实验室实测的土壤含盐量,从相关系数、标准差及信息熵三个角度探讨分数阶微分算法对土壤高光谱数据的影响。结果表明:随着微分阶数的增加,相关系数通过0.01显著性检验的波段数量总体上呈逐渐减少的趋势,且1 lg R提升相关性的效果优于其他三种数学变换;高光谱数据总体分布变得相对集中,样本差异性逐渐降低;信息熵逐渐减小,信息无序度变小,有效信息量增加。分数阶微分能够细化相关系数、标准差及信息熵的变化趋势,丰富高光谱数据的预处理方法,可从光谱维的角度深层挖掘光谱信息,为深度利用高光谱数据提供崭新的视角,同时也可为特征波段选择、地表参数反演等高光谱数据的应用提供参考依据。 Taking the Aibi Lake basin as the research area and the typical saline soil as the research object, the fractional differential is introduced. Taking 0.2 as the differential order interval, 0 ~ 2 is subdivided into the 11th order differential. The original spectral reflectance and its common Root mean square, reciprocal and other mathematical transformation differential calculation, combined with laboratory measurements of soil salinity, from the correlation coefficient, standard deviation and information entropy from the perspective of fractional differential algorithm on soil hyperspectral data. The results show that as the order of differentiation increases, the number of bands through which the correlation coefficient passes the significance test decreases gradually, and the effect of 1 lg R improves the correlation better than the other three mathematical transformations. The total hyperspectral data The distribution becomes relatively concentrated, the sample difference gradually decreases; the information entropy decreases gradually, the information disorder degree becomes smaller and the effective information amount increases. Fractional differential can refine the trend of correlation coefficient, standard deviation and information entropy, and enrich the pretreatment method of hyperspectral data, which can deeply explore the spectral information from the perspective of spectral dimension and provide a brand new perspective for the further utilization of hyperspectral data, meanwhile, It also provides a reference for the application of hyperspectral data such as feature band selection and surface parameter inversion.