L-AVI: A Dual-Head Illumination-Aware Adaptive Vision Framework for Extreme Lunar High Dynamic Range Environments

L-AVI (Lunar Adaptive Vision Intelligence) is a dual-head, illumination-aware adaptive perception framework designed to address extreme high dynamic range (HDR) conditions encountered in permanently shadowed regions (PSRs) and sunlit lunar terrains. Future sustained lunar operations demand robust visual autonomy capable of functioning across abrupt illumination transitions, low albedo surfaces, and severe contrast gradients. The proposed architecture integrates a shared convolutional encoder with two specialized processing heads: (1) an illumination enhancement branch optimized for shadowed and low-signal regions, and (2) a structural reconstruction branch focused on edge preservation and terrain consistency. A dynamic illumination-aware gating mechanism enables adaptive feature fusion based on local radiometric characteristics. To quantitatively evaluate performance under simulated lunar HDR conditions, the framework is assessed using Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index (SSIM) metrics. Experimental results demonstrate a PSNR of 22.08 dB and an SSIM of 0.5060 under extreme contrast scenarios. Statistical validation is reported using mean ± standard deviation analysis across test subsets. The system is designed as a TRL-4 level laboratory-validated prototype, targeting future integration into autonomous navigation stacks for lunar rovers, robotic prospecting systems, and sustained surface infrastructure operations. This work contributes toward the development of illumination-robust perception systems capable of supporting long-duration extraterrestrial missions under severe photometric variability.