 The proposed work presents a novel classification framework for Paul SAR images using compact and adaptive implementation of CNNs with sliding window approach, which has advantages such as no requirement for extensive feature extraction, computational efficiency, and ability to perform classification using smaller window sizes than deep CNNs. Experimental evaluations on four benchmark Paul SAR images show overall accuracies ranging between 92.33 to 99.39%. This article was authored by Mita Hishali, Serkan Kiranias, Tarkoints, and others.