A Scattering Physics-Driven Network for Damage Detection of Scattering Targets

Diagnosing damage of scattering targets with multiple fault sources is highly challenging, mainly because scattering-source reconstruction is ill-posed, insufficient precision and inadequate physical interpretability. To address this challenge, a scattering physics-driven two-stage network (SP-TN) is proposed for target damage detection. The core novelty involves: by employing the mapping relationship in SAR imaging, a physics-driven solution for inverse scattering problems is established; integrated with physical constraints on location and structure, the underlying physical rules of damage manifestations are traced, enabling the precise scattering diagnosis. The damage detection datasets containing 1856 high quality SAR images have been constructed to study the effects of single-point damage and multiple-point damage on feature extraction and model detection accuracy. The scattering target testing and ablation experiments demonstrate that the SP-TN achieved high-precision localization, classification, and scattering entropy analysis for single-point and multiple-point scattering damage faults with 0.92 precision, providing a novel approach for scattering damage diagnosis.