Research On Core Algorithms Of Onboard Hyperspectral Image Analysis System

Hyperspectral has drawn a lot of attention for its powerful ability to detect different materials. But the mass data, produced by hyperspectral sensor, has brought heavy burden on the transmission and storage system of the satellite. And the gap between data generating rates and transfer rates will become larger in future. At the same time, not all data are valuable. These useless data waste the insufficient onboard resource. So, it's necessary to develop onboard processing technology in order to select valuable information from the hyperspectral image intelligently.This thesis focuses on the core algorithms of onboard hyperspectral image analysis system. A framework of onboard hyperspectral image analysis system has been proposed at first. Then the core algorithms, spectral unmixing and target detection, have been developed to support the performance of the system.A new algorithm for spectral unmixing has been developed. The algorithm, which is based on linear mixing model, contains two procedures: prediction and adjustment. Prediction generates a set of endmembers by using projection. With the information provided by the pixel outlied the simplex formed by predicted endmembers, adjustment expands the simplex to make the estimated endmembers closer to the true endmembers. In a series of experiments using simulated and real data, the algorithm competes with VCA, with better performance.A subpixel target detection algorithm based on compressed data has been developed. The algorithm, derived from Adaptive Matched Subspace Detector, detects subpixel target using endmembers and abundance maps produced by spectral unmixing. It largely decreases the computing complexity, while preserves the performance of the original algorithm. Compare to the other algorithm based on compressed data, the algorithm is more robust on low signal-to-noise rate (SNR).