Hyperspectral Images Based On Projection Pursuit Feature Extraction And Automatic Identification Technology Research,

Hyperspectral remote sensing is an art, which integrates the spectrum representing to the radiant attributes of ground object with the homological images standing for spatial and geometric relations. For it has an extremely high spectral resolution, such data has facilities to the physicochemical characteristics mining or subtle recognition of different ground objects. In fact, traditional panchromatic and multi-spectral imagery processing algorithms have been not to fulfill the needs for extracting information from hyperspectral imagery, as it have different structure characteristics and data features. How to extract interest information from massive spectral imagery data automatically and effectively is a challenge in hyperspectral remote sensing applications research.Hyperspectral imagery feature extraction and classification is approached by combining with nation's 863-708 project and corresponding scientific assignments. It focused on the automated ground object recognition. Hyperspectral imagery feature selection, extraction and automated classification are researched at large. Meanwhile, the essential idea, implementing method and algorithm about projection and pursuit is introduced. Finally, a projection pursuit-based (PP) scenario for hyperspectral imagery feature extraction and recognition is proposed. In summary, following the mainly researches are carried out in this dissertation:1. The essential idea, implementing method and algorithm about projection pursuit are introduced; the projection index (PI) and optimal searching algorithms are also summarized and analyzed.2. The hyperspectral imagery feature extraction and recognition are researched at large, and some common feature extraction algorithms are introduced. The hyperspectral imagery feature extraction solution to address the hyperspectral data characters adopted sequential projection pursuit (SPP) is proposed. following are these advantages: The redundancy information is wiped off, the data dimensions is diminished, and the data processing/analysis capabilities are improved remarkably in virtue of correlation coefficient between spectral bands as feature selection criteria. Hughes phenomenon from insufficiently training sample is restrained effectively through the hyperspectral imagery feature dimensions diminishment adopted SPP, and the operational efficiency is numerously...