Target recognition with information entropy based multi-task sparse representation in SAR imagery

Weak relatedness among tasks leads to failure of regularized multi-task sparse representation (RMTSR) model to handle target recognition in synthetic aperture radar (SAR) imagery. Therefore, it is vital to measure task relationship not only in order to obtain desired model but shrink the size of dictionary and the training time. In this paper, sparse representation under each feature modality is considered as a single task in RMTSR. A nonlinear sparsity correlation index (NSCI) is presented. Furthermore, nonlinear correlation information entropy (NCIE) deduced from NSCI is utilized to quantify the relatedness among tasks from view of information theory. Experiments conducted on MSTAR demonstrate the outperformance and effectiveness of RMTSR even in the case of limited training resource. Moreover, NCIE is efficient to measure the generalization of model and select appropriate feature set to reduce complexity.