基于精化成像模型的高光谱卫星三维融合图像生成方法

In recent years, the optical satellite 3D fusion image generation becomes a new research frontier with the rapid development of hyperspectral imaging and stereo imaging technology. The main problems of optical satellite 3D fusion image generation are as follows: Due to the great resolution difference between panchromatic and hyperspectral imagery, the object contour cannot be aligned together, so that leads to the edge blur and spectral distortion of fused images; The low accuracy of relative orientation of stereo images leads to the failure of the binocular disparity between epi-polar image pairs. As a consequence, to achieve the high-fidelity fusion of panchromatic and hyperspectral imagery, the project builds a novel fusion model based on multiplicative transformation; Moreover, the binocular parallax image generation method for stereo images based on refined imaging model is explored to pave the way for the extensively use of 3D fused image in map applications and image interpretation.

近年来，高光谱成像和立体成像技术的快速发展使光学卫星三维融合图像生成研究成为一个新前沿领域。目前，三维融合图像生成主要存在以下问题：全色与高光谱图像分辨率差异大，目标轮廓无法套合，导致融合图像出现严重的边缘模糊和光谱失真现象；光学卫星立体像对的相对定向精度较低导致核线图像不能形成双目视差。为此本项目建立基于乘性变换的新融合模型，实现全色与高光谱图像的高保真融合；同时，研究基于精化成像模型的立体像对双目视差图像生成方法，为三维融合影像在地图应用和影像判读中应用奠定技术基础。

当前光学卫星大多数采用全色与多光谱/高光谱传感器同时相成像方式，而图谱合一的融合图像是卫星应用的基础，是战术目标识别的支撑技术。目前其他遥感图像融合技术存在光谱和细节失真问题，多年来未得到解决，严重影响了遥感图像的准确分析与应用。实际应用需要高保真融合全色、多光谱/高光谱图像，才能实现卫星应用由“黑白”时代进入“彩色”时代。本项目揭示了加性变换模型产生失真的机理缺限，发现了空间细节锐化因子及其求解方法，构造了乘性变换融合模型，提出了基于精化成像模型的立体像对双目视差图像生成方法，并证明其能达到高保真融合效果，实现了机理突破。