A new era of Synthetic Aperture Radar (SAR) has begun in recent years. Convolutional neural networks (CNNs) have garnered a lot of interest lately for their ability to analyze SAR data. This paper thoroughly studies the main subfields of SAR data analysis that CNNs have addressed, including segmentation, change detection, object identification, automatic target recognition, land use and land cover classification, and image denoising. Particular attention has been paid to useful methods like transfer learning and data augmentation. To overcome the issues of a high false alarm rate and the challenges of attaining high-performance detection using traditional approaches, a deep learning-based SAR target identification and classification method is proposed for target detection tasks in complex backdrops. An optical based approach is presented in this study for enhancing the security feature. A machine learning-based radar with better performance is suggested in light of the deep learning-based target models' problems with a high parameter count and memory usage. Even though there have been some significant advancements, deep learning research in radar is still mainly being tested in lab settings and is still in its theoretical stage. There are still a number of obstacles and potential restrictions in the application, including issues with dataset adequacy, robustness of the model, and electromagnetic modelling fidelity. Nonetheless, it is undeniable that deep learning technology will significantly advance radar. As a result, it is wise to recognize the field's current difficulties and potential future paths. Furthermore, it is hoped that this review will give readers fresh opportunities to investigate appropriate deep learning-based methods for radar applications. In this study, Long Short-Term Memory (LSTM) and SqueezeNet model are used for enhancing the accuracy of system designed.