Fractional Fourier Transform Based Joint Adaptive Subspace Detection for Hyperspectral Anomaly Detection

The use of a transform domain is usually helpful to improve the performance of hyperspectral anomaly detection (AD). By fractional Fourier transform (FrFT), the signals obtain complementary characteristics in an intermediate domain between the original reflectance spectrum and its Fourier transform. In this letter, a joint adaptive subspace detection (ASD) model based on FrFT (FJASD) is proposed for hyperspectral AD. The FJASD has two parts. First, hyperspectral imagery (HSI) is transformed by FrFT and background joint sparse representation (BJSR) that is the ASD model based on the test point vector is used in the fractional Fourier domain (FrFD) for AD. This part mainly uses spectral characteristics of HSI. Then, the tensor block center at the test point in the original domain is transformed into a 1-D vector and dimensionally reduced by a principal component analysis (PCA). The lower dimensional dataset is transformed by FrFT and tensor-based ASD (TBASD) that is the ASD model based on the test point tensor is used in FrFD for AD. This part mainly uses the spatial characteristics of HSI. In addition, the spectral information after removing the redundancy is also utilized. Finally, the above two parts are combined by a weight coefficient and complement each other. Four comparison algorithms are used to verify the AD performance of FJASD through three hyperspectral datasets and the detection results show the superiority of the proposed FJASD.

Automated summary

The paper proposes a joint adaptive subspace detection model based on fractional Fourier transform (FJASD) for hyperspectral anomaly detection. The FJASD combines spectral and spatial characteristics to improve the performance of AD.