Inverse Scattering–Driven Variational Reconstruction With Morphology-Aware Regularisation for Wax-Covered Pest Imaging

Wax-mediated surface layers introduce anisotropic diffusion and non-Lambertian reflectance that distort observable pest morphology in agricultural imagery, degrading the reliability of data-driven detection models under field conditions. A physically constrained imaging framework is proposed based on a spatially varying scattering operator parameterised by diffusion length, anisotropy, and dominant orientation, capturing directionally dependent degradation while remaining bounded and interpretable. Reconstruction is formulated as a variational inverse problem combining data fidelity with morphology-aware regularisation, integrating edge-preserving gradients and curvature-based constraints to enforce anatomical consistency. The resulting optimisation problem is solved using an ADMM-based scheme with a linearised treatment of the curvature term, which converges reliably to a stationary point in practice. A local perturbation analysis gives a conditional bound on sensitivity to additive observation noise, while robustness to $\pm 15\%$ perturbation of the scattering parameters is assessed experimentally. On synthetic data the method improves PSNR by 5.2 dB, SSIM by 0.15 and boundary F-score by 26% relative to total-variation deconvolution, and remains 2.0 dB and 0.03 SSIM ahead of the strongest learning-based baseline evaluated. Semi-synthetic and field experiments reproduce the same ordering, with expert agreement rising from 0.58 to 0.78 and contrast recovery from 0.70 to 0.84 under realistic imaging conditions and without scene-specific tuning. Beyond pest imaging, the proposed formulation addresses a broader class of inverse imaging problems involving spatially varying anisotropic scattering, where conventional convolutional and learning-based models are less well suited.