Key Technology Research On Multidimensional Information Fusion Based On Polarization Spectral Images

Polarization spectral imaging data has multi-dimensional information such as spatial,intensity,spectral and polarized information compared to traditional spectral data.Traditional camouflage targets are similar in color to the natural background,and the detection in the spectral dimension is susceptible to interference from changes in external factors,prone to the phenomenon of "foreign objects with the same spectrum",resulting in a single spectral information is difficult to meet the needs of accurate identification of camouflage targets.The multi-dimensional characteristics of the polarization spectrum can reflect more information about the inherent properties of the target,and depending on the surface roughness and material,different polarization difference characteristics can be generated.The polarization information image reflects the detailed information such as the outline and texture of the object,which can,to a certain extent,make up for the shortcomings of low contrast and insufficient information in the spectral image,and has advantages for the detection and identification of spectral camouflage targets in complex backgrounds.In this paper,we study the imaging mechanism and data characteristics of polarization spectral data,and acquire polarization spectral data in multiple scenes based on the principle prototype of the self-researched polarization spectral imager.In this paper,we design a waveband reconstruction and polarization feature acquisition algorithm for the data structure and characteristics of the polarization spectral imager,and propose a bilateral filtering image pre-processing method to improve the quality of the polarization spectral data,which realizes the pre-processing and information reconstruction of the original data of the polarization spectral images.For polarization spectral image fusion,this paper introduces the traditional multiscale decomposition method,proposes a fusion method of local binary mode features based on non-downsampled contourlet transform,uses local binary mode features for weighted fusion of high-pass sub-bands according to the characteristics of polarization feature images with small grey scale values and influenced by illumination.We conduct a comprehensive evaluation of the experimental results,It is shown that the method proposed in this paper effectively achieves polarization feature fusion and enhances the difference between the artificial target and the natural background.A comparison of the target recognition accuracy is also carried out by combining the plain image,the polarization fusion image and the polarization hyperspectral image,and the results show that the polarization hyperspectral fusion image effectively improves the recognition accuracy of the camouflaged targets.In terms of visual significance enhancement,a method based on the similarity measure of visual significance is designed to carry out multi-scale fusion of different waveband information,and then false-color fusion of the polarimetric parameters after the fusion of each waveband.The experimental results show that the method designed in this paper can effectively enhance the visual effect,enhance the polarimetric contrast of the target,and the fused images have significantly improved in terms of information entropy,contrast,average gradient and other evaluation indexes.