Hyperspectral anomaly detection with local correlation fractional Fourier transform and vector pulse coupled neural network

Abstract

Hyperspectral anomaly detection has attracted widespread attention because of its significant practical application value. However, the existence of noise and the lack of target spatial shape and texture features in hyperspectral image (HSI) limit the performance of hyperspectral anomaly detection. Here, we present a novel hyperspectral anomaly detection method by local correlation fractional Fourier transform (FrFT) and vector pulse coupled neural network (VPCNN). First, combined with the correlation between local pixels, the HSI is transformed into the fractional Fourier domain through the proposed local correlation FrFT, which can suppress noise and background while preserving anomaly targets features in the HSI. Second, the VPCNN, which can effectively deal with the lack of spatial shape and texture features of the target, is proposed to segment the transformed HSI into different regions to obtain the segmentation image. Finally, anomaly detection result is extracted from the segmentation image. Experimental results on four real-world datasets show that the proposed method has better superiority and advancement compared with other state-of-the-art alternative methods. We hope that our research will inspire the application of segmentation methods in related fields.