Rotated Detection of SAR Ship Target based on Angle Prediction of Continuous Boundary

In deep learning-based rotated synthetic aperture radar (SAR) target detection, the boundary discontinuity of angle prediction is a key problem that affects the prediction accuracy. Since the convolutional neural network (CNN) is a continuous function, the output of the function is continuous, but the boundary of the angle prediction is discontinuous, so it is impossible to use continuous CNN to fit the discontinuous angle prediction. Based on this, this paper proposes to use phase-shifting coder (PSC) to encode the angle. The PSC itself can be predicted using continuous CNNs, so the PSC coding can alleviate the boundary discontinuity problem to a large extent. Thus, accurate and continuous angle prediction can be achieved by using CNNs. The experiment results on the HRSIC dataset also show the effectiveness of the proposed method.

I. Introduction

In the field of remote sensing, synthetic aperture radar has all-weather, all-day imaging capability [1] . Ship detection is a crucial application of SAR imagery, holding significant importance in both military and civilian sectors. Accurate location data offers vital intelligence for governments and military forces to swiftly address maritime emergencies [2]. With advancements in artificial intelligence, ship detection techniques utilizing convolutional neural networks (CNNs) have demonstrated outstanding performance [3].