Noise Reduction-Based Image Enhancement Using Transformer Networks for Satellite SAR Radar Recognition

In this paper, we propose a method for improving image transformer recognition based on noise removal for satellite SAR (Synthetic Aperture Radar) radar. Recently, the U.S. military uses SAR radar data to observe and track conditions on land or at sea without being affected by weather conditions. However, recognition techniques that are widely used in existing SAR radars have limitations in the recognition rate. In particular, image transformer techniques such as Vision Transformer (ViT) have recently been widely used in the field of computer vision, but research on target recognition using SAR radar results is insufficient. In addition, when the effect of noise is large due to various environmental conditions, SAR radar needs various attempts to increase the recognition rate of image transformer techniques. To solve this problem, this paper discusses how to improve Wavelet-based image transformer recognition. This method improves the recognition rate by removing noise before applying the Transformer technique to SAR radar data. Applying the proposed method, we show an approximately 87% improvement in the recognition rate for ViT compared to the previous one.