Automated classification of land-covers from satellite images using machine learning and deep learning approaches have garnered significant attention among researchers in the computer vision field in recent years due to the real-life significance of this application. Among various types of land covers mangroves stand out as a unique ecosystem that plays a crucial role in regulating the climate in coastal areas and providing a habitat for diverse flora and fauna. These characteristics make mangrove vegetation particularly significant. In this work, we present a novel and computationally efficient deep learning architecture called MangNet, designed to effectively classify diverse land covers in the Sunderban region of West Bengal, India. MangNet exploits its multi-scale contextual feature extraction capability to simultaneously classify multiple pixels within a patch, considering the interrelationship between neighboring pixels, which enhances its performance efficiency. The designed architecture achieved an impressive classification accuracy of 99.64%, surpassing existing state-of-the-art methods. We have used freely accessible dual-polarized Sentinel-1 data from the chosen Region of Interest (RoI) to conduct this study and demonstrated the relevance of intensity and phase values in each polarized channel for classification in the Results section.