Hyperspectral Anomaly Detection via Sparsity of Core Tensor under Gradient Domain

Hyperspectral anomaly detection (AD) task is a typical binary classification problem, and utilizing background prior knowledge key technique to solving such problems. The two most commonly used priors for hyperspectral images are low-rank local smooth properties. Most traditional matrix-based methods use regularizations model these types of integrate them into one model, which makes unable maximize their effectiveness. In addition, the matrix method also destroys structure (HSI). To address issues, this study identified unique sparsity property in gradient tensor HSI. Specifically, core resulting from Tucker decomposition was observed exhibit sparsity. This property, referred as GCS (the on map), effectively captures structural information HSI improves performance. regularization offers following advantages: 1) uses term simultaneously capture both low-rankness smoothness, size represents background, ℓ 1 norm describes map, i.e., smoothness original data; 2) constrained regularization, allowing full utilization different dimensions when updating tensor, spatial spectral carried by three-factor matrices decomposition. Finally, extensive experiments validate superiority our proposed methods.