Surface Scene Simulation Of Full Spectral Imager

Image simulation is to acquire simulated remote sensing images via digital models,based on understanding and analyzing of imaging process.Nowadays,the theory of image simulation has been widely applied in various territories of remote sensing,among which more significant application is to provide important references for orbiting test and development of satellite,so its reliabilities of designment and function can be ensured.With the development of image simulation,image simulation techniques have flourished in narrow spectral range,such as visible or thermal infrared band.Corresponding study in full spectral range is still inadequate,however,because of complicated radiative transfer process and big spectral range.In the paper,based on the above problem,the research of surface scene simulation is done in full spectral range.Based on the orbiting test of GF-4 satellite and the development of GF-5,the paper studies image simulation via correlations in neighbor bands.GF-4 was launched in Xi Chang Satellite Launch Center on December 29,2015,which is the first remote sensing satellite of geosynchronous orbits in china.To ensure the normal work of GF-4,the study uses HJ satellite data to carry out surface scene simulation in visible and near infrared bands.GF-5,equipped with high spectral camera and several detectors of atmospheric environment and ingredient,is to plan to be launched in 2017.The paper employs the data of Aster to do surface scene simulation in full spectral range of GF-5 to provide important references for prelaunch design.The studied band scope in the paper is full spectral range,including visible and nearby infrared,short wave infrared,mid-wave infrared and thermal infrared.Based on the research in published paper,a method of surface scene simulation based on spectral correlation in neighbor bands is proposed.This method first considers spectral correlation between bands of data source and objective data and then resamples spectral information via spectral response function of remote sensing sensor.Convert model between real data and objective data band is constructed through Partial Least Squares Regression and evaluated by 10-fold-cross-validation.The model which has the least error is the ultimately convert model in neighbor bands.Then convert model is applied on image of data source to generate simulated remote sensing image.The proposed method is applied on surface scene simulation of GF-4 and GF-5,and then simulated images are acquired.Structural similarity is used to assess similarities between simulated and actual images.As a result,the simulation in visible and near infrared bands of GF-4 has less 5%of average relative errors and 69%?85%of similarity;and the simulation in full spectrum of GF-5 has 10%of average relative errors.