Maritime Target Detection Based on Multi-Domain Feature Refinement Fusion Network

Due to the complex and changeable characteristics of the sea state, maritime radar target detection has always been a challenging problem. Recently, deep learning-based maritime target detection has promoted detection accuracy to a certain extent. However, most deep learning-based methods tend to only focus on single-domain features, so that the generalization ability and robustness of these methods are still not ideal. To this end, this paper proposes a new target detection method named multi-domain feature refinement fusion network to achieve high-precision maritime radar target detection. First, the signal is encoded into time-frequency graphs and Gramian Angular Field graphs. The features extracted from the two domains can provide complementary information to more fully characterize the target information in the echoes. Then, a multi-domain feature fusion network is developed, which appeals to the idea of cross-attention to realize multi-domain feature fusion while solving the problem of image misalignment caused by cross-domain feature fusion. The evaluation experiments on the measured IPIX dataset demonstrate the proposed method outperforms several state-of-the-art deep learning-based target detection methods.