Research On Compression Of Ultraspectral Infrared Atmosphric Image Based On Thinning And Prediction

Ultraspectral atmospheric sounding provides important observationnal data for applications and research in the field of meteorology.With the improvement of spectral detection technology,the spectral resolution is greatly improved and the revisit period is shortened.Thus,observationnal data containing more abundant and fine information of atmospheric state parameters can be obtained by the ultraspectral sounder.However,thounands of channels and shorter observation period have brought a sharp increase in the amount of data.Therefore,research on effective data compression methods is of great significance to the transmission and storage of ultraspectral sounder data.This paper mainly studies the lossless compression method for the ultraspectral sounder data obtained by a single observation for a certain area and method for data sequence obtained by multiple observations for a certain area.Characteristics of ultalspectral sounder data including the spatial-spectral-temporal correlation,the segmentation characteristics based on the physical properties of the spectrum and statistical characteristics of emissivity data value are analyzed to demonstrate data redundancy and its compressibility,identifing the basis for latter study of compression methods.The method based on key information extraction and spatial-spectral prediction is used to compress the ultraspectral sounder data obtained by a single observation.This method focuses on how to extract the key information that meets the application requirements by radiance thinning,and how to predict all data based on the key information,so that lossless compression can be achieved by transmitting key information,predicting residual errors and other necessary parameters.In the method,in order to achieve spectral thinning to obtain key channels including sensitive channels and auxiliary channels,temperature and humidity sensitive channels are selected by a method of stepwise iteration based on information content,and a part of auxiliary channels are selected using the correlation-based spectral clustering method.The data of key channels is then spatially thinned to obtain the key information,which is decorrelated by a PCA-based method before being encoded and also spatially interpolated to obtain the prediction residual errors of the data before spatial thinning.The data of remaining channels is predicted by the auxiliary channels based on results of spectral clustering to obtain the difference data for the remaining channels.Later,the difference data is randomly undersampled according to the theory of compressed sensing and sequently reconstructed after sampling.The generating residual errors of reconstruction,the undersampled observation data,the dictionary of the difference data and the measurement matrix will transmitted to ensure the lossless compression of the difference data of the remaining channels.Compression experiments has been performed on 10 orbits of IASI L1 C data,achieving the experimental compression ratio varing from1.89 to 2.01.For the ultraspectral sounder data obtained by multiple observations,a compression method based temporally online prediction using data characteristics of physical segmentation is studied and ultlized in this paper.According to the physical segmentation characteristics and the temporal correlation of ultraspectal sounder data,a temporally bidirectional prediction model based on spectral segmentation is established,where online learning is used to update the prediction parameters of the model.Compression experiments have been performed on the 6 data cubic acquired by the sounder of GIIRS within 12 hours,with the highest compression ratio of 1.90.