Multiobjective Sparse Subpixel Mapping for Remote Sensing Imagery

摘要: Subpixel mapping (SPM) of remote sensing imagery is aimed at generating a classification map with finer spatial resolution based on the abundance maps. The sparse subpixel (SSM) method reformulates SPM problem into pattern linear regression preconstructed patch dictionary. However, in SSM model, optimization ${L}0$ -norm nonconvex NP-hard problem, so ${L}1$ used to replace obtain an approximate solution, and selection optimal weight parameter between multiple terms difficult. Thus, this paper, novel multiobjective (MOSSM) framework for proposed, which transforms problem. In MOSSM, first, sparsity term accurately modeled using instead avoid potential errors caused by -norm, evolutionary algorithm directly optimize -norm. Second, subfitness-based employed simultaneously fidelity term, prior generate set coefficients balance these three terms. there no need determine sensitive parameters. Finally, two terms, can be applied overcomplete dictionary, are presented proposed MOSSM-TV MOSSM-L algorithms incorporate correlation subpixels. Experiments were conducted synthetic images real data sets, results compared those ten other demonstrate effectiveness method.