Hyperspectral Anomaly Detection Based on Fractional Domain Sensing and Contrastive Statistical Regularized Networks

Hyperspectral anomaly detection is a highly challenging task, primarily because weak anomalous targets within high-dimensional spectral data are easily obscured by complex and heterogeneous backgrounds. However, existing methods still have limitations. They typically rely on strong background assumptions or focus solely on spectral or spatial features, neglecting frequency-domain information. This makes it difficult to effectively distinguish subtle anomalies in complex scenes. Therefore, this paper proposes a method based on fractional domain sensing and contrastive statistical regularized network (FDCRN), which adopts the diffusion model as the basic framework for background suppression and optimizes anomaly detection performance from three dimensions including frequency domain enhancement, spatio-spectral fusion, and distribution correction. First, we design a lightweight fractional Fourier transform (LFrFT) module, which adaptively modules the spectral frequency components through learnable fractional-order parameters and combines frequency domain features to enhance the capture of subtle differences of weak and small targets. Second, we propose a multi-scale spatial-spectral enhancement (MSSE) module, which extracts multi-scale spatial information by using depth-separable convolution and fuses spectral features to enhance the spatial-spectral correlation of weak and small targets. Third, we introduce a statistical adaptive and contrastive learning (SACL) module to dynamically correct the feature distribution shift and increase the semantic distance between the background and the anomaly. Experiments on multiple real hyperspectral datasets show that FDCRN outperforms existing methods in detection accuracy, response intensity to weak and small targets, and adaptability to complex scenes, providing a more efficient solution for hyperspectral anomaly detection in complex scenarios.