A Generalized New Method for Anomalous Phased Array Radar Echo Image Restoration Based on Generative Adversarial Network

Abstract X‐band Phased array radars are characterized by high spatial and temporal resolution, but suffer from a range of data quality problems, such as echo voids after the filtering of ground clutter, abnormal radials, radial obstructions and irregular missing radar echoes. This paper proposes a radar echo image restoration model (GCD) based on color correction and detail enhancement of adversarial generative networks with dual‐stream encoder‐decoder. To overcome the challenge of restoring strong echo regions, a multi‐scale Feature Alignment (FA)‐based strong echo color correction module is designed to achieve FA between original features and radial obstruction features. Additionally, a Local Detail Enhancement Module is designed to enhance the high‐frequency texture information in the strong echo regions. The GCD not only applies to all types of radar PPI images but also significantly reduces the time required for initial data processing compared to traditional methods (Sliding Window Filling method). Specifically, the speedup ratio for single‐image testing is 361.28. Addressing the issue of radial obstructions, GCD achieves an improvement of 10.15 in PSNR and a notable decrease of 18.921 dBZ in |dBZ|, thereby resolving the problem of false enhancement of meteorological echo edges that occurs with the Sliding Window Filling method. When compared to the base model, GCD further enhances the restoration of strong echoes, with a decrease in |dBZ| by 0.654. The method proposed fills the gap in the field of radar data quality control in image processing and promotes the development of artificial intelligence in this area.