Research On Key Technologies Of Geostationary Orbit High-Resolution Meteorological Satellite Image Processing

Remote sensing satellites are an important platform for obtaining ground object information.Its image products are widely used in many fields such as weather forecasting,resource surveys,urban planning,disaster monitoring,environmental protection and military surveys.In the field of meteorological applications,meteorological remote sensing satellites have gradually developed into a joint observation system with polar orbit and geostationary orbit.Compared with polar orbiting satellites,geostationary orbiting satellites can conduct high-frequency observations in the same area,to improve the time resolution of observations.However,affected by the factors such as working environment,imaging equipment and imaging methods,the data observed by the multi-channel scanning payload carried on the geostationary orbit remote sensing satellite will be degraded during imaging,transmission,and storage.In other words,the observation data needs to be preprocessed before application.This dissertation restores and reconstructs the degraded data,and explores the four main technologies: 1)the misalignment between adjacent rows caused by the bidirectional scanning mode(Intra-frame registration);2)the change of the channel-to-channel co-registration error caused by factors such as temperature field(Channel-to-channel co-registration);3)Improving images spatial resolution on the basis of existing hardware(Super-resolution reconstruction);4)the string noise caused by the non-uniformity of the detector pixel response(Remove noise).Data degradation will directly affect the accuracy of the interpretation,information extraction,and subsequent applications of remote sensing images.Therefore,this dissertation relies on the advanced geosynchronous radiation imager(AGRI)to focus on the key technologies such as denoising,registration,and reconstruction of meteorological remote sensing satellite images for the above four problems.The following innovative research work has been carried out:(1)Aiming at the misalignment between adjacent rows,a mathematical model based on the Fourier phase shift theory and payload scanning mode is established.The model first calculates the phase difference spectrum curve of adjacent pixels between adjacent misaligned rows.Secondly,the displacement value is obtained by fitting the low frequency part of the phase difference spectrum by using the least squares method.Finally,the dislocation rows are reconstructed by using bilinear interpolation according to the misalignment value to solve the misalignment.The algorithm can achieve registration between adjacent rows with high precision on the sub-pixel level.(2)Aiming at the calculation of the channel-to-channel co-registration,combined with the characteristics of AGRI meteorological satellite images,a joint strategy of two co-registration methods is proposed,and can systematically calculate co-registration errors between image pairs.For channels with a clear boundary between ground objects and cloud information,a conventional method based on normalized mutual information and parabolic interpolation is used to calculate the channel-to-channel co-registration errors.For infrared channels with blurred boundary between ground objects and cloud information,a method based on the phase correlation that regards the extracted cloud information distribution image as the co-registration source is used to calculate the channel-to-channel co-registration errors.Finally,the co-registration errors of all channels are completely calculated by the combination of the two methods.The proposed algorithm processes AGRI on-orbit images.The results show that the algorithm has better robustness,real-time performance,and stability and can achieve channel-to-channel co-registration with high precision.(3)Aiming at the infrared channel of AGRI with sub-pixel imaging technology,a multi-frame super-resolution reconstruction method based on the maximum a posteriori probability(MAP)framework is proposed.In this dissertation,the degradation model is decoupled according to the imaging mode.First,the phase correlation method is used to check the co-registration error.Then,according to the relationship of blur function between the low-resolution image and the reconstructed image,the blur function used in the degraded model is estimated from the low-resolution images.After estimating the main parameters of the super-resolution image reconstruction model,this dissertation chooses a regularized reconstruction method based on the MAP framework.In the fidelity term,the L2 norm corresponding to the Gaussian noise model is used,and in the regularization term,the bilateral total variation(BTV)with strong edge protection capability is used.The reconstruction results of simulated images and AGRI original images show that the super-resolution reconstruction method in this dissertation can effectively improve the image spatial resolution.(4)Aiming at the problem of stripe noise caused by the non-uniformity noise,the mathematical optimization model is proposed based on the scanning mode of the payload and stripe noise structural properties.The stripe noise image is estimated from the observed image to achieve the purpose of denoising.Unlike most existing stripe removal optimization models,in the method proposed in this dissertation,the regularization based on the L1 norm represents the global sparseness of the stripe image structure;the constraint condition based on difference is used to describe the smoothness of the stripe direction(scan direction)and discontinuity of stripes in the vertical direction(stepper direction).To protect the image’s detailed structural information,this dissertation introduces an edge weight factor to the constraint item in the stripes’ vertical direction.Finally,the proposed optimization model is solved and optimized by the alternating direction multiplier method(ADMM).By processing the AGRI on-orbit data and comparing it with the typical destriping noise method,the results show that the proposed algorithm can better retain the detailed information while eliminating the stripe noise and better qualitative and quantitative results.