Research on Maritime Target Radar Detection Network

Abstract

A Range-Doppler map, which can be obtained from radar received signal, allows for the estimation of a target’s range and speed. In maritime military applications, the Range-Doppler map is used to detect hostile targets such as warships and missiles. Since the presence of sea clutter and target signal’s sidelobes increases the probability of false alarms, necessitating robust detection method against the sea clutter and side- lobes. Therefore, this study proposes a AI-based maritime target detection method using the characteristic of the target’s Doppler spectrum. The deep learning-based ap- proach has the advantage of not requiring manual hyperparameters tuning by humans unlike traditional signal processing methods, which require setting hyperparameters for each dataset. A maritime environment simulator is designed based on a existing research and generate military data of various scenarios. Using the data, a robust mar- itime target detection network that performs well even in unfamiliar environments is designed and proposed. In the proposed network, the Line Proposal Unit, typically used for lane detection, is appropriately modified and additional network module is designed to estimate the target’s pixel within the Range-Doppler map. While the pro- posed network demonstrates superior detection performance for high-speed targets compared to traditional methods, its performance for low-speed targets deteriorates. Furthermore, the detection performance for multiple targets also shows degradation. Therefore, future research must address these issues by modifying the deep learning network’s concept and structure and the loss function.