AeroLLE: Constrained Pseudo-Supervision for Nighttime Aerial Image Enhancement with the AeroNight-1.5K Benchmark

2026-08-01

AI Summary

This study addresses the challenges of nighttime aerial image enhancement, which are commonly affected by non-uniform exposure, mixed illumination, and weak structural information, compounded by the absence of strictly aligned ground-truth reference images under normal lighting. To this end, the authors introduce AeroNight-1.5K, the first benchmark dataset comprising 1,500 real-world samples, supporting both paired and unpaired evaluation protocols. They further propose AeroLLE, a two-stage enhancement framework: an initial HVI-based enhancer restores visibility, followed by a Spatially Adaptive Exposure-Color Calibration (SAECC) module that enables pseudo-supervised optimization without requiring registered references, preserving geometry and texture fidelity. By imposing constraints on correction magnitude and spatial variation, the method outperforms existing approaches in visual consistency, exposure balance, and color accuracy, demonstrating the efficacy of staged, constraint-aware calibration.

Abstract

Nighttime aerial image enhancement is challenged by spatially nonuniform exposure, mixed illumination, and weak structural evidence, while registered normal-light targets are difficult to capture from moving platforms. Generated normal-light images provide practical appearance guidance but may alter geometry or texture. We introduce \aeronight{}, comprising 1,500 real nighttime aerial RGB images: 1,300 inputs are associated with manually screened pseudo-references, and 200 inputs support unpaired evaluation. We propose AeroLLE, a two-stage framework that first recovers visibility with an HVI Base Enhancer and then performs Spatially Adaptive Exposure--Color Calibration (SAECC). After the Base Enhancer is selected and frozen, SAECC predicts bounded, low-resolution RGB gain and bias fields, restricting the magnitude and spatial variation of the second-stage correction. Experiments under complementary pseudo-paired and unpaired protocols demonstrate improved agreement with screened appearance targets, together with more balanced exposure and color correction across diverse nighttime aerial scenes. These results support constrained, stage-specific calibration as a practical strategy for learning from generated appearance guidance when registered aerial references are unavailable.

Problem

• nighttime aerial image enhancement
• nonuniform exposure
• mixed illumination
• pseudo-supervision
• appearance guidance

Innovation

• nighttime aerial image enhancement
• pseudo-supervision
• spatially adaptive calibration
• AeroNight-1.5K benchmark
• constrained enhancement