Multi-Feature joint weighted hyperspectral target detection based on optimized Gabor filter

Abstract

With the rapid development of hyperspectral remote sensing technology, target detection technology exhibits a wide range of application prospects in many fields. However, the detection algorithms based on raw spectral features suffer from single feature extraction and insufficient joint spatial-spectral information, leading to their poor detection performance in complex scenes. To address these issues, this paper proposes a multi-feature joint weighted hyperspectral target detection based on optimized Gabor filter. First, the conventional 3D Gabor filter is optimized through spectral angle similarity measurement, which adaptively enhances responses in target feature regions. Second, amplitude and phase information are innovatively combined, and representative feature cubes are screened based on four optimization criteria, effectively reducing data redundancy while improving feature representation capability. Finally, a weighted fusion strategy is used to integrate the four preliminary detection maps and iteratively optimized with the fuzzy Otsu algorithm, which significantly improves the detection accuracy and algorithm stability. The method proposed in this paper achieves outstanding detection performance on six hyperspectral datasets.