Adverse-Weather Aerial Object Detection: From Degradation Mechanisms to Robustness Evidence

Adverse weather degrades visibility, contrast, edges, and local target details in aerial imagery, while small object size, viewing geometry, and onboard constraints further amplify detection failures. This review examines how weather-induced degradation propagates from image formation to feature damage and detection errors, and evaluates the evidence supporting existing robustness claims. A structured search of studies published between 2015 and 2026 identified 174 candidate records, which were analysed in terms of degradation mechanisms, methodological pathways, datasets, evaluation protocols, training dependencies, and deployment conditions. Existing methods are grouped into input restoration, joint restoration–detection learning, degradation-aware representation, domain adaptation and generalisation, multimodal and temporal compensation, and lightweight deployment. The evidence shows that improved visual quality does not consistently translate into better detection performance, while many reported gains depend on paired data, target-domain access, external priors, additional sensors, or computational overhead. Current evidence remains strongly concentrated on haze, with limited validation for natural rain, snow, dust, compound weather, unseen-weather generalisation, uncertainty, and closed-loop flight. Future research should therefore prioritise realistic weather data, controlled evaluation, risk-aware assessment, and onboard validation.