Land Use Classification with Sparse Models

This chapter aims to tackle the problem of land use (LU) image classification. The task of classification consists of assigning labels to fixed-size test images based on a ground truth training set. The chapter describes the two fusion strategies, namely gray image-based feature fusion and spectral fusion. Thus, the issue is addressed from two perspectives. The first one is articulated over fusing several feature vectors extracted from the input image, whereas the second one consists in fusing feature vectors extracted from the inherent spectral layers of the input image. The chapter explains compressive sampling (CS) or sparse sampling theory and the adopted reconstruction technique and presents experimental results and interpretations. Orthogonal matching pursuit (OMP) and stagewise orthogonal matching pursuit (StOMP) algorithms are used to solve optimization problems. In contrast to the basic OMP algorithm, StOMP involves many coefficients at each stage, whereas in OMP only one coefficient can be involved.