EGAD: an edge-guided attention dehazing network for aircraft landing views
Yueting Chen, Yichun Tai, Fan Dong Meng, Zhijiang Zhang · Journal of Electronic Imaging · 2025
Compared with daytime flights with optimal visibility, images captured during the aircraft’s landing phase under hazy conditions exhibit significant degradation in quality, thereby increasing the risk of runway incursions, whereas the current dehazing algorithm performs poorly on the airport runway dataset and fails to meet the specific requirement of clearly capturing foreground information. To address the above issue, considering the changing perspective of aircraft landing, we propose an edge-guided attention network for airport runway image dehazing (EGAD), with the objective of preserving runway marking line textures and enhancing the overall dehazing image quality. Specifically, the proposed EGAD consists of a dehazing branch and an edge extraction branch, integrating features generated by them via a process-oriented approach. The edge extraction branch serves as an additional edge prior, supplemented by edge loss to impose quadratic constraints on the network. In addition, the dehazing branch incorporates multiple enhanced local attention modules to enhance detail extraction capabilities. Moreover, we synthesize a runway dataset tailored for aircraft landing scenarios. To our knowledge, we are the first to address the challenging task of dehazing aircraft landing perspective images. Extensive experiments on the runway dataset validate that our EGAD significantly outperforms state-of-the-art methods in terms of both visual quality and quantitative metrics, and the detection algorithm achieved higher detection accuracy on the images after dehazing.