Comparative study on brightness temperature reconstruction of synthetic aperture based on deep learning

In order to study the effect of deep learning method in the brightness temperature reconstruction of synthetic aperture radiometer, the gray value of remote sensing image with channel amplitude-phase error and random error is used as the original brightness temperature image for simulation experiment to compare the brightness temperature reconstruction images under traditional Fourier transform, CNN inversion, U-net inversion and Resnet inversion. From the view of image visual effect, Resnet inversion method has the best image restoration effect and the weakest background noise. From the evaluation index, the RMSE of Resnet inversion method is the smallest, which is 6.28K, and the PNSR value is the highest, which is 31.62dB. The second is CNN inversion method, RMSE value is 10.93K, PNSR value is 27.09dB. Therefore, Resnet inversion method can better restore bright temperature image, reduce Gibbs effect and improve image resolution.