Improved classification of conservation tillage practices using hyperspectral imagery with spatial-spectral features

Classification of conservation tillage practices from hyper-spectral imagery is challenging due to spectral similarity between soils and senescent crop residues. In this paper, a novel classifier using both spectral and spatial information is proposed for hyperspectral image classification. Three steps are included: (1) a feature extraction method using a very simple local averaging filter to produce the joint spectral-spatial features; (2) an efficient local Fisher discriminant analysis projection for dimensionality reduction and class separability enhancement; and (3) the typical k-nearest neighbor classifier for final classification. Experimental results using real hy-perspectral data demonstrate the benefits of the proposed approach, which can outperform other popular classifiers, such as support vector machine with composite kernel.