Hyperspectral Image Classification of Agricultural Tillage Practices Using Spatial-aware Collaborative Representation

Classification of conservation tillage practices from hyperspectral imagery is challenging due to spectral similarity between soils and senescent crop residues. Using both spectral and spatial information can greatly improve the classification accuracy. In this paper, we investigate the collaborative representation classifiers in tillage mapping due to their highly efficient performance for large-scale image classification, which is because such classifiers do not require a training phase. To better utilize the spatial information in addition to the spectral information in a hyperspectral image, we focus on a spatial-aware collaborative representation classifier, which can directly incorporate the spatial information during classification by adding a spatial regularization term to the objective function. Experimental results demonstrate its higher accuracy and lower computational cost compared to other types of classifiers.