Aperture Synthetic And Mirror Aperture Synthetic Radiometer Deep Learning Image Reconstruction Method

In the field of microwave radiation measurement in geostationary orbit,both aperture synthetic radiometer(ASR)and mirror aperture synthetic radiometer(MASR)have application prospects.In principle,both ASR and MASR measure the spatial frequency domain data of the observed scene.This frequency domain data need brightness temperature image reconstruction to obtain the brightness temperature image of the observation scene.However,in the actual system of brightness temperature image reconstruction,whether ASR and MASR,existing brightness temperature image reconstruction algorithms have their own shortcomings.In order to improve the quality of brightness temperature image reconstruction,this paper proposes a corresponding method for brightness temperature image reconstruction based on the data and error characteristics of ASR and MASR.The main innovative research work of this paper is as follows:(1)Based on the analysis of frequency truncation in uniform sampling ASR,a uniform sampling ASR image reconstruction method(ASR-T method)based on Transformer depth learning network is proposed.This method uses network to learn the mapping relationships in brightness temperature image reconstruction.The simulation and experimental results show that the image quality reconstructed by the ASR-T method is better than the existing methods.(2)The method of non-uniform sampling ASR brightness temperature image reconstruction(IASR-T method)is proposed.This method adds frequency domain recovery and other functions on the ASR-T method.In loss function,brightness temperature image and corresponding spectrum are considered to optimize network parameters.The simulation and experimental results show that the IASR-T method outperforms existing methods in terms of image reconstruction quality.(3)Based on the analysis of the rank-deficiency error caused by the vertical MASR,a vertical MASR brightness temperature image reconstruction method based on deep convolution neural network(CNN)is proposed(MAS-CNN method).The input data of this method is the correlation matrix,which avoids the rank-deficiency error introduced by solving the visibility function through the correlation matrix.In the network training process,the channel amplitude and phase errors of the system are considered,and the network learns the error information to improve the quality of image reconstruction.The simulation and experimental results show that the MAS-CNN method outperforms existing methods in terms of image reconstruction quality.(4)Based on the analysis of the basic principle of tilted MASR and the reflector error,a method of tilted MASR brightness temperature image reconstruction(MAS-SET method)combining Transformer and CNN is proposed.Firstly,the method extracts global and local feature information from the correlation matrix,and then fuses the extracted feature information through a specially designed network structure to reconstruct the brightness temperature image.Through simulation,it is demonstrated that the MAS-SET method is superior to the impulse matrix reconstruction method in terms of image reconstruction quality.In the experiment,the measured tilted MASR data are obtained through multiple experiments,and the ASR-T method is introduced to obtain the training label,and the measured training dataset is generated for the MAS-SET network training.The trained MAS-SET network is used for brightness temperature image reconstruction of measured data,and the quality of the brightness temperature image obtained is better than that of the impulse matrix reconstruction method.