Two-dimensional nearest neighbor classification for agricultural remote sensing

Two-dimensional nearest neighbor classification algorithm (2DNNC) is proposed for analyzing agriculture remote sensing data by combining matrix feature leaning and matrix-based dictionary learning. In the framework of 2DNNC, all hyperspectral feature vectors are transformed into matrix features by a set of nearest neighbor classifiers. The matrix features contain significantly discriminant information because of the label information from the set of nearest neighbor classifiers. Taking advantage of these matrix features, discriminant matrix dictionary is learned for classification by rank-1 matrices approximation. Experimental results on agriculture remote sensing data show the effectiveness and efficiency of the proposed algorithm on hyperspectral image classification.