Infrared polarimetric imaging enhancement via multiscale inverse modulation and bio-inspired neural reconstruction

Infrared polarization image enhancement and reconstruction are crucial for national defense applications such as target recognition, environmental perception, and weak target detection. Single-modality infrared polarization imaging, however, faces key challenges: limited polarization feature extraction, insufficient background suppression, and unstable reconstruction quality. In the 2024 Anhui field experiment, conventional DoLP (Degree of Linear Polarization) models showed high false-detection rates in low-polarization regions. To address this, we propose a maximum response channel difference representation, improving spatial frequency by up to 30%, and a pixel-level inverse modulation with dynamic weighting, achieving up to 370% contrast enhancement under occlusion. An NSST-MSD-PCNN framework further integrates shearlet-based geometric sensitivity with adaptive pulse-coupled neural networks. Validation on the IPEG (Infrared Polarization Evaluation Gallery) benchmark and deployment on the DJI M300 RTK UAV demonstrate superior performance in both subjective and objective metrics, providing an engineering-ready solution for robust infrared polarization imaging in complex environments.