新型超高分辨率星载多光谱与全色光图像融合方法研究

With the rapid development of sensor techniques, some new features can be found in the new satellite remote sensing images. These features, such as the high spatial resolution, the more spectral bands, the wider spctrum coverage, represent the trend of the upcoming remote sensing images and bring new challenges to image fusion field. Hence, WorldView-2 satellite images are accepted as the experimental data in this project. Usually, the pansharpening process consists of two stages: upscaling and fusing. This research work will include (1) the high-fidelity interpolation model for multi-band data and (2) the effective fusion model based on correspondence analysis. In the first part, the N-band multispectral images will be treated as a whole data set and the interpolation model combines both the spatial and the spectral relationship between the interpolated and original locations in order to preserve more spectral and spatial information during the upscaling stage. The second part puts emphasis on building the effective fusion model. Band selection based matching model will be investigated according to the relative spectral responses between multispectral and panchromatic images. Then the low frequency component of panchromatic image will be accurately estimated, and the spatial details (high frequency component) can be extacted from the panchromatic image. To prevent the distortions caused by the injecting process, proper injection models should be studied and help inject the details into the component image produced by correspondence analysis. Finally, the effective pansharpening method for the new very high resolution satellite images will be presented. The aforementioned research will be valuable for pansharpening the new remote sensing images, such as WorldView-2, WorldView-3, etc. The findings can be helpful to many remote sensing applications, such as high precision detection, classification and recognition of earth objects, urban planning, disaster evaluations, etc. This project has great significance for science research and broad application prospects.

随着传感器技术的快速发展，新型星载遥感图像具有空间分辨率更高、波段数更多、光谱覆盖范围更宽等新特点，代表了下一代遥感图像的发展趋势，同时也对现有的图像融合技术提出挑战。本项目采用最新型的WorldView-2星载遥感数据为实验对象，研究在多光谱图像重采样环节进行整体插值以提高重采样数据的光谱和空间信息保真能力；从多光谱与全色光传感器的光谱响应特点出发，研究基于波段选择的光谱匹配模型，准确估计全色光图像的低频分量，从而建立全色光图像空间信息有效提取方法；利用基于对应分析的强度分量构建方法进一步研究空间信息的高保真注入模型，以减少信息在注入过程中导致的光谱与空间失真现象，在此基础上研究适用于新型超高分辨率星载遥感图像的图像融合方法，为新一代遥感图像的融合技术奠定理论方法基础。研究成果可用于高精度地物检测、分类与识别、城市规划、灾难检测与评估等领域，具有重要的科学研究意义和广阔的应用前景。

随着传感器技术的快速发展，新型星载遥感图像具有空间分辨率更高、波段数更多、光谱覆盖范围更宽等新特点，代表了下一代遥感图像的发展趋势，同时也对现有的遥感图像融合技术提出挑战。本项采用WorldView-2卫星数据为研究对象，对新型高分辨率星载多光谱与全色光图像的融合方法开展了深入研究，研究内容包括1）建立了多光谱图像的整体插值方法。该方法兼顾各光谱矢量的相互作用和像素点的空间特性，减少了重采样放大环节引入的空间和光谱信息失真，可以同时对任意波段图像数据进行整体插值放大。2）提出了基于对应分析的图像融合方法。该方法设计构造出合适的列联表，在成分空间得到成分分量，可以有效地分离出任意波段多光谱数据的强度分量和光谱信息，有助于融合质量的改进。3）提出了基于波段分组的图像融合方法。该方法分组策略简单而有效，突破了现有融合方法对波段间光谱响应匹配关系的过分依赖，提高了融合的质量。4）提出了几种基于导向滤波的融合方法。利用导向滤波器的保边平滑和结构转移特性设计融合模型，不仅能有效去除重采样图像的块效应，还可以准确提取全色光图像的空间细节信息。提出的方法不受多光谱图像输入波段数的限制，既可以对单一多光谱波段直接实施融合，也可以对多波段图像整体融合处理，适用于高光谱图像。项目研究成果为新一代遥感图像的融合技术奠定了理论方法和研究基础，可服务于高精度地物检测、分类与识别、城市规划、灾后评估、GIS系统等领域，具有重要的科学价值和广阔的应用前景。