Multi-task Joint Sparse and Low-rank Representation Target Detection for Hyperspectral Image

Target detection plays an important role in hyperspectral imagery (HSI) processing. Many detection algorithms have been proposed over the past decades. However, the existing detectors may encounter false alarms for ignoring target interference during background modeling and high correlations among adjacent bands. To address the target interference issue, we propose a novel joint-sparse and low-rank representation target detection algorithm for HSI, which separately models target and background pixels using different regularization methods. A background pixel in HSI can be modeled via sparse and low-rank representation using a background dictionary, whereas a target pixel can be modeled via sparse representation using a target dictionary. To reduce spectral redundancy, we further incorporated the detection model into a multitask learning framework. The final detection was made in favor of the class with the lowest total reconstruction error accumulated from all tasks. Experiments on two airborne HSIs demonstrated that multitask joint-sparse and low-rank representation (MTJSLR) outperformed other state-of-the-art detectors.