Active Learning Using Multifeature Kernel Sparse Representation Classifier for Hyperspectral Image Classification

Active learning is a popular technique that mitigates the need for large training samples for hyperspectral image classification. The success of this technique is dependent on the query function designed to select informative samples for labelling. In the literature, most of the query functions are defined by exploiting spectral information. In this research, a novel query function for active learning is defined that considers both the spectral and spatial information of the hyperspectral image using multiple feature sets. The multiple feature sets provided complementary information for selecting those informative samples which are not only informative in spectral perspective but also from spatial perspective. Experiments conducted using two real hyperspectral datasets show that the proposed technique outperformed many state-of-the-art techniques.