A Multidimensional Information Fusion Algorithm for Polarization Spectrum Reconstruction Based on Nonsubsampled Contourlet Transform
Jing-jing Zhong,Xiao Liu,Xue-ji Wang,Jia-cheng Liu,Hong Liu,Chen Qi,Yu-yang Liu,Tao Yu
DOI: https://doi.org/10.3964/j.issn.1000-0593(2023)04-1254-08
2023-01-01
Spectroscopy and spectral analysis
Abstract:This paper proposes a polarization spectral multidimensional information fusion method based on nonsubsampled contourlet transform to address the shortcomings of traditional optical methods that make it difficult to identify camouflaged spectral targets in complex backgrounds and the common fusion methods that tend to lead to image information loss. The multidimensional information reconstruction algorithm was designed based on the acquired multidimensional information such as target space, spectrum and polarization, and the basic data of polarization state including Stokes parameters as well as the degree of polarization and angle of polarization were extracted. NSCT is used to fuse the basic polarization parameters to improve the image's information content and improve the camouflage's recognition accuracy. The images Q and U with the same edge information are first decomposed using NSCT. Regional energy-weighted fusion is performed for the decomposed low-pass sub-bands; for the high-pass sub-bands, LBP features are used for weighted fusion according to the characteristics of polarization features, such as small gray value and high influence by illumination. At the same time, the proposed method is compared with four types of fusion methods, and the fusion results are evaluated objectively according to five indicators: information entropy, standard deviation, mean gradient, contrast and peak signal-to-noise ratio, and the target recognition accuracy is compared with plain images, polarized fused images and polarized hyper-spectral images. The information entropy of the fused image is 6.998 6, the standard deviation is 45.599 8, and the average gradient is 19.808 6. Compared with the original intensity, the improvements are 5.1%, 14.04%, and 7.3%, respectively, ranking first among the four types of fusion methods. It is shown that the method proposed in this paper effectively achieves polarization-based feature fusion and enhances the difference between the artificial target and the natural background. At the same time, the recognition accuracy of the fused polarized hyperspectral image for the target reaches 0.986 2, which is 21% higher than the target recognition accuracy of the single-intensity image. The experimental results show that the proposed method can effectively fuse the intensity and polarization information to improve image contrast and readability. The fused image also significantly improves target recognition accuracy, overcoming the problem of high false alarm rate of traditional spectral means for camouflage target recognition, and providing a new and effective means for new concept spectral camouflage disclosure, which has great application value.