Target Detection in Clutter/Interference Regions Based on Deep Feature Fusion for HFSWR

High-frequency surface wave radar (HFSWR) is of great significance for maritime detection, but in the HFSWR echo signal, ship targets are often submerged in a variety of clutter and interference, making it difficult to detect vessels. In this paper, we propose an intelligent detection algorithm for targets concealed in strong clutter and complex interference environments. The algorithm has two stages: preprocessing and target detection. In the preprocessing stage, faster region-based convolutional neural networks Faster R-CNN are designed to identify and locate clutter and interference regions in the range Doppler spectrum; in the target detection stage, a two-level cascade algorithm is proposed. First, an extremum detection algorithm is proposed to identify suspicious target points in the clutter/interference regions, including real and false target points, to quickly obtain potential target positions. Second, in consideration of the characteristics of radar targets, two lightweight networks are designed to extract the CNN features and the stacked autoencoder features of the potential target locations. Then, fusion features are obtained and sent to an extreme learning machine that acts as a second-level classifier to distinguish between real and false target points. Experiments show that the proposed HFSWR target-detection algorithm has better performance for vessel detection in clutter/interference regions than the current mainstream detection algorithms.