Differentiable scattering model for cone-shaped targets and its application in scattering center modelling

The scattering center (SC) model, though widely used in neural networks, often relies on high-frequency asymptotic approximations that introduce singularities and non-differentiability, hindering its compatibility with gradient-based optimization. In this paper, we propose a method to eliminate these singularities by introducing the Gamma function, and construct a differentiable scattering model with respect to both frequency and aspect angle. This model effectively suppresses Doppler spectrum leakage and mitigates the issue of gradient explosion in neural networks. Its accuracy in simulating radar cross-section (RCS) fluctuations and Doppler frequency characteristics is validated through full-wave numerical computations. Furthermore, the model is embedded into a physics-informed neural network (PINN) for parameter inversion, demonstrating its strong performance in neural network learning tasks.