Two inverse processes: Spectral reconstruction and pixel unmixing

The application of hyperspectral remote sensing has been a research focus in recent years, and one of its fundamental goals is to detect and classify the constituent materials for each pixel in the scene, which is pixel unmixing. This research proposes spectral reconstruction, which is the inverse process of pixel unmixing. It can be used to analyze the essence of the hyperspectral imaging procedure and provide inverse analysis to the pixel unmixing. Also, it may find the physical origin of hyperspectral sensors' parameters degenerating with the stability of multispectral scanner, which helps to provide quantitative basis for the development and improvement of hyperspectral sensors. Based on research for years, we can get hyperspectral data of high quality and stability from multispectral data, which is a new way to the hyperspectral application. The technical route is using the normalized multiple endmember decomposition method (NMEDM) to decompose the endmember data of vegetation, water and soil based on the condition of fuzzy sets and full constraint. The characteristics are as follows: this new method considers the space-time variation of the pixel endmember, the hyperspectral reconstruction can be achieved with less calculated amount, and the terrain spectral reflectance can be showed with less data. This research includes: deeper mining of the multispectral information, sensitivity analysis of multispectral bands, verification of the hyperspectral reconstruction model based on the ground multispectral imaging.