Research On Sub-pixel Mapping Of Hyperspectral Imagery Based On Spectral Unmixing And Objective Optimization

Due to its rich spectral information, hyperspectral images are widely applied in many fields, becoming one of the most important source of information on Earth observation. However, due to the limit of imaging principles and manufacturing technology, the spatial resolution of hyperspectral image is relatively low, mixed pixels are pervasive in the images. For land cover mapping, coastline extraction, change detection, landscape indices estimated and so on, spatial detail data in the mixed pixels is of critical importance. It is not correct to judge mixed pixels to be any category by the traditional hard classification. Sub-pixel mapping is an effective means to solve the problem. Therefore, the research of sub-pixel mapping technology is of great significance.This article studied on the sub-pixel mapping of hyperspectral images based on the combination of hyperspectral image spectral unmixing with the intelligent optimization algorithm. The main works of this article are as follows:(1) The first part introduces the background and significance of the research. Summarize and analyze the relevant literatures, which provide an important theoretical basis for further study of the sub-pixel mapping.(2) The second chapter briefly introduces the basis theories of spectral unmixing, including the definition of spectral unmixing, and mathematical model, typical methods of spectral unmixing with the pure pixel assumptions, finally introduce typical methods of spectral unmixing without the pure pixel assumptions.(3) The algorithm framework of sub-pixel mapping based on spectral unmixing and objective optimization; determine the minimum perimeter of image connected area as the objective function, and introduce three different image perimeter calculation methods, further analyze the applicable to optimization algorithm of sub-pixel mapping. In order to decrease the running time of the algorithm, local analysis is used to replace the global analysis according to the feature space distribution characteristics, propose anew iteration strategy of objective optimization.(4) Introduce the principle of genetic algorithm and binary particle swarm optimization algorithm, as well as the details of the application of these two algorithms in sub-pixel mapping, including the representation of a particle and update process. Combining with three different calculation methods of objective function, compared the two kinds of optimization algorithm in the application of sub-pixel mapping.(5) By analyzing the impact on perimeter and region number, which is caused by some special cases such as isolated point or region including only two points, the regional perimeter is modified and the cost function is formulated by considering regional perimeter as well as the number of connected regions.(6) The last chapter summaries the whole work of the paper and put forward the future work on hyperspectral images sub-pixel mapping.