Research On Spectral Calibration And Image Stitch Based On Push-broom Multi-module Imaging Spectrometer

Hyperspectral remote sensing technology gets both spatial and spectrum information of target using dozens or hundreds of narrow and continuous spectrum channels. Compared to multispectral technology, hyperspectral technology provides a fine spectral curve of target. Therefore it is widely applied in target identification, mineral exploration, agriculture assessment, disaster warning and so on.In this paper, based on ―Wide-band imaging spectrometer‖ project and its indicator, the overall design program were studied and several system performance test has been conducted. In these experiments, improving the performance of spectral property and multi-module image stitching is the research focus. To solve the problems about the two points, we wrote the software for test, studied related theory and proposed improvement program.In order to improve the quantum efficiency of the spectrometer, back-illuminated thinned type CCD was chosen, but it brings up interference fringes which reduce the spectral resolution in near-infrared band. To solve this problem, we established a multi-beam interference model similar to a Farby-Pérot interferometer, verified its correctness with measured data, and analyzed the relationship between CCD photosensitive zone thickness and the interference phenomenon. On this base, we use improved flat-field correction algorithm to correct the interference. In addition, the interference fringes are sensitive to the distribution of incident light wavelength and stable in spatial frequency, so it is very suitable for the measurement and calibration of spectral deviation. In actual experiment, the interference model was used to estimate the distribution of stripes, and then using frequency domain filtering and least squares fitting methods to extract the phase information of interference fringes. These information were used as an additional parameter for spectrum calibration, which can reduce the dependence on the light stability of calibration system effectively, and improve detecting accuracy for slight offset in spectral dimension.In order to achieve multi-module sub-pixel image stitching efficiently and exactly, the method which dose pixel match at first and then dose sub-pixel matching is presented. First, we briefly introduce the current image registration algorithms based on gray value and feature extraction, and compare the two methods in different application conditions. Then, considering the strengths and weaknesses of the existing sub-pixel registration algorithms, mutual information and PV interpolation methods are chosen to calculate the relationship between images to be matched. At last, for the defects of PV interpolation algorithm that local optima always appear near image pixels, two different methods are proposed to improve. Experiment results show that the algorithm can calculate the transformation between images token from different modules effectively.