Hyperspectral anomaly detection using Taylor expansion and weighted irregular block filter

초록

Hyperspectral anomaly detection (HAD) is a fundamental task in remote sensing image processing to identify anomalous targets that differ from the background spectrum. However, the interference of local background in hyperspectral images (HSIs) can lead to high false alarm rates in the detection results. We proposed a novel HAD method based on Taylor expansion and weighted irregular block filter (WIBF) to address this issue. First, to the best of our knowledge, we first introduced a single-pixel-based Taylor expansion method into the field of HAD to capture the different spectral features of the HSI. Second, a three-dimensional (3D) hyperspectral feature pyramid was constructed to extracted the feature image of the enhanced image. Third, a WIBF method was proposed to filter the feature image to suppress local background and preserve anomaly information. Fourth, an adaptive dual weight fusion method was proposed for the weighted fusion of the energy images to suppress noise and background. Finally, anomaly detection results were extracted from the reconstructed HSI, which was obtained by the inverse Taylor expansion. Experiments conducted on synthetic and real-world datasets demonstrated the effectiveness and superiority of the proposed method.

키워드