Polarized hyperspectral image fusion method for targets in sea clutter background

Abstract

Maritime target detection in complex marine environments is challenging, particularly with sea clutter interference, which hampers the efficacy of traditional radar and optical spectral or polarization features in target discrimination. To address this, a radiant energy model is developed to simulate the polarized hyperspectral properties of the ocean, forming the basis for the proposed detection method. This study introduces a spectral-polarization feature fusion method to enhance target detection accuracy by integrating hyperspectral spectral information with polarization feature structure. The first-order difference method is employed to compute the gradient of linear polarization. For each pixel, the cosine of the average gradient direction across all bands is calculated and used as a weighting factor. The intensity image is then weighted and fused, followed by principal component analysis (PCA) to extract the most representative bands. Experimental evaluations under varying signal-to-noise ratios show that the fused image significantly improves contrast and standard deviation metrics. Specifically, the fused image achieves 7.64 % and 9.41 % improvements in contrast and standard deviation, respectively. Moreover, the fused polarized hyperspectral image attains 98.91 % target recognition accuracy, a 12.24 % improvement over the polarized multispectral fused image. In terms of detection performance, it also yields notable gains in precision by 11.2 %, recall by 8.6 %, and F1-score by 11.4 %. The proposed method effectively suppresses sea clutter and enhances target detection accuracy and robustness.