Oveis et al. majority of previous SAR studies that deal with CNNs are classification tasks and only a few studies have extended and applied CNNs for regression-type problems. Note that classification and regression problems with CNN do not necessarily have different networks. In fact, learning classification problems are also solved through regression and then the result is discretized. Here the main difference is the choice of the loss functions. In regression problems, mean squared error (MSE), root MSE (L2), mean absolute error (L1), and Huber loss are typically used, whereas, in