Intelligent cascaded HF radar maritime target detection via Background-Matched Subspace Projection and Improved Dual-Scale Branch Fusion Network

High-frequency surface wave radar (HFSWR), as an important sensor for maritime surveillance and measurement, is a critical asset for coastline maritime surveillance and over-the-horizon target detection. To address the challenge of ship detection with HFSWR in heterogeneous clutter, this paper proposes a cascaded detection framework integrating Background-Matched Subspace Projection (BMSP) with an Improved Dual-Scale Branch Fusion Network (IDSBFN). First, BMSP enhances target-clutter separability by projecting whitened snapshots onto local background subspaces under varying rank hypotheses, constructing a 3D “Range-Doppler-Rank” tensor that highlights target scattering features. Subsequently, a normalized energy detector performs coarse detection, and the resulting candidates are fed into the IDSBFN for fine classification. The network features a dual-branch architecture that extracts global and local features, adaptively integrating them via a gating mechanism, while a multi-head self-attention module refines cross-channel textural representations. Finally, an adaptive thresholding strategy is established by modeling the Pareto optimal front between the first-stage preset false alarm rate and the second-stage classification threshold. Experimental results on measured HFSWR data demonstrate that the proposed method significantly enhances detectability limits for weak targets and exhibits increased robustness in heterogeneous clutter formed by the interplay of sea and ionospheric interference, thereby improving the overall measurement robustness of radar systems in complex and dynamic maritime environments.