Pixel Level And Feature Level Remote Sensing Image Fusion Methods And Applications

The development of sensor technology enriched human’s accesses to information, and remote sensing technology nowadays has already become the most important means for human to get information about the earth. Satellite remote sensing systems provide information rich and diverse types of remote sensing image data for the application of earth observation and the study of geosciences. Using image fusion technologies to merge different types of remote sensing images and to accurately and efficiently extract useful information from these images has become a key issue in applications of remote sensing technology. Therefore, we launched a study on the methods and related theories of remote sensing image fusion.The research described in this thesis mainly contains the following three aspects:1, A novel pansharpening fusion method has been proposed aiming at resolution enhancement of multi-spectral remote sensing images. Improvement of image quality is the main concern of pixel-level image fusion, and spatial resolution is the most important quality of remote sensing images. Because the radiation energy captured by the sensors is limited and the observations are usually interferenced by noises, the qualities of high spatial resolution and high spectral resolution can hardly be achieved at the same time in remote sensing images. However, using pansharpening technologies to fuse multi-spectral images with panchromatic image, synthetic images with both high spatial and spectral resolution can be obtained. In order to obtain a pansharpening method with outstanding fusion performance, an elaborately designed color space transform is employed. This color space transform is a standard orthogonal transform based on the linear regression of image data. Furthermore, the idea of multi-resolution analysis is also applied to complete the construction of the fusion method. The superiority of the proposed method has been verified in comparative experiments.2, A thermal sharpening method has been proposed to achieve resolution enhancement of thermal infrared remote sensing images. Thermal infrared images provide information on surface temperature, which is critical in quantitative remote sensing applications, therefore the research of thermal sharpening methods is practically meaningful. Thermal sharpening is achieved on the fusion of thermal infrared image and visual near-infrared images, and due to the different characteristics of these two kinds of remote sensing images, common fusion methods of pixel level can not be used to implement the fusion. On the other hand, how to make full use of the spacial details contained in the multi-channel visual near-infrared images is another essential issue for thermal sharpening. In this thesis, a high-speed neural network algorithm is adopted to establish a regression model as the core structure of the fusion method for thermal sharpening. The efficiency of the proposed thermal sharpening fusion method has been shown in experiments using actual remote sensing data. 3, A feature level remote sensing image fusion method has been proposed to conduct quantitative analysis of surface evapotranspiration information. Pixel level image fusion is the process of upgrading the quality of the image data, while feature level image fusion is the process of extracting information by the integration of multiple remote sensing images. Quantitative analysis of surface information, including evapotranspiration related information, is an important issue in the research of remote sensing technology. A number of intermediate parameters are involved in the process of quantifying surface evapotranspiration, therefore, the subject can be solved through a complex multi-step fusion procedure. The fusion rules are established based on the ground surface structure model and the physical relationship between surface parameters. Feature fusion results are compared with the surface measured data to prove the validity of the proposed method, and a comprehensive understanding of the state of the studying area can be obtained according to these feature fusion results.