A dimension reduction model for sparse hyperspectral target detection with weighted ℓ1 minimization

Target detection for hyperspectral image is an important application in many given areas. As natural signals can always be represented sparsely using a given dictionary, some effective sparse-based algorithms are developed for target detection. However, this kind of method is extremely dependent on the spectra library, a huge dictionary data set from which useful information is extracted. This provides mathematical challenge for its efficiency of programming. Under such circumstances, we propose a novel algorithm called dimension reduction based sparse detector (DRSD), which aims to eliminate the calculation process by building a smaller new spectra library. In addition, as the detection result is based on the sparse reconstruction the algorithm provide, we utilize effective approximate solver to find sparse reconstruction within the new spectra library. The experimental results demonstrate that the proposed algorithm is more effective than current applied sparse algorithms.