A Nonnegative Sparse and Collaborative Method for Hyperspectral Target Detection

Algorithms based on sparse representation (SR), such as sparsity-based target detection (STD) and sparse representation-based binary hypothesis (SRBBH) model for target detection, have attracted widespread attention in the field of hyperspectral target detection. A key point of the SR-based target detection methods is the representation or recovery of the pixel to be detected. However, few studies focus on this problem, they simply use the orthogonal matching pursuit (OMP) to implement the representation. To improve this situation, this paper proposes a nonnegative, sparse, and weighted collaborative method with 10 norm (NSWCL0). Taking into account the various characteristics of the coefficients in the sparse representation, NSWCL0 remodels the recovery process. Using the alternating direction multiplier method (ADMM), we achieved the solution of the model. By simply adopting the framework of STD and SRBBH, we can accomplish hyperspectral target detection. The proposed hyperspectral target detection algorithms have been tested on one synthetic and two real hyperspectral images. The experimental results show that the algorithms applying the proposed recovery method have better detection performance than other algorithms.