Hyperspectral soil texture classification

Abstract

A soil texture classification system is developed and exploited in the hyperspectral domain. The hyperspectral signatures of three different pure soil textures, i.e., sand, silt and clay, combined with a linear mixture model, are used to generate signals representing different types of soil textures. Feature extraction via the discrete wavelet transform and linear discriminant analysis for feature vector reduction and optimization are used. Different types of classifiers, which include the nearest mean and maximum likelihood, are incorporated to test the system's applicability. Classification accuracy is evaluated using a leave-one-out method. Experimental results are presented and possible future works are discussed.