Hybrid Neural Network-Based Coherent Detection Framework for Low-Altitude Maneuvering Target Over Sea-Surface

Coherent detection typically relies on multi-pulses coherent integration (CI) to achieve the effective target energy focusing, so that it can be highlighted and detected from the background. However, for the low altitude maneuvering target over sea-surface, the range migration (RM), Doppler migration (DM) and severe sea clutter often occur simultaneously in the echo, which seriously affect the coherent detection performance. To solve these problems, this paper proposes a coherent detection framework for low altitude maneuvering target over sea-surface based on hybrid neural network. This framework can be divided into two main steps: CI based on CV-HUNet and clutter suppression based on Dual-ResNet structure. First, the CV-HUNet is applied to extract the target energy trajectory, and the motion parameters of the maneuvering target are estimated by polynomial fitting. Based on the estimation results, the RM and DM can be compensated to obtain the CI results. Second, the CI results are input into the Dual-ResNet structure to classify the target and clutter components and then complete clutter suppression. After the above two procedures, the processing outputs are used for constant false alarm rate detection(CFAR). The publicly measured datasets collected by Naval Aviation University (NAU) of China in 2021 are applied to compare the coherent detection performance of the existed typical methods with the proposed one, which have demonstrated the validation of the given framework.