Colourization of SAR Image Using Generative Adversarial Network

Abstract-Employing generative adversarial networks, specifically with regard to cycle consistency loss and mask vectors, mainly concentrates on the colorization of Synthetic Aperture Radar (SAR). Most SAR imagery is devoid of chromatic information. Contemporary deep learning techniques are the predominant approach for SAR colorization. The methodology proposed herein employs a multidomain cycle-consistency generative adversarial network (MC-GAN). It enhances performance through the integration of a mask vector and cycle-consistency loss. The approach does not necessitate the availability of paired SAR-optical imagery. The multidomain classification loss contributes to the precision of the color output. The methodology has been evaluated using the SEN1-2 dataset for urban and terrain areas.