A New SAR Autofocus Method Based on Error Coefficient Optimization

To address the problem of blurred Synthetic Aperture Radar (SAR) imaging caused by motion errors, this paper proposes a new SAR autofocus method based on the traditional Minimum Entropy Autofocus (MEA) algorithm. First, we model the phase error in SAR images as a polynomial error and divide the error characteristics into high-order error and low-order error. Then, we propose a Restormer and Convolutional Neural Network (CNN) Model (RCM) to estimate the error coefficient, where the feature extraction network based on Restormer is used to predict the low-order error coefficient and the feature extraction network based on CNN is used to predict the high-order error coefficient. Finally, we compensate the predicted phase error in the original SAR image to obtain the focused SAR image. The experimental results show that the SAR focused image obtained by our proposed method has better results in entropy and contrast metrics.