An Efficient Dual-Branch Guided Nighttime Image Dynamic Range Enhancement Network for Remote Airport Tower Control

In the field of remote airport tower control, the captured images often suffer from unsatisfactory limited dynamic range at night, which presents challenges in ensuring accurate airport management. To meet the requirements on both dynamic range enhancement and processing speed, an efficient dual-branch guided nighttime image dynamic range enhancement network is proposed for airport high-resolution images. Firstly, the YUV color space was utilized to decouple luminance and chrominance from RGB channels, thereby mitigating instability during the enhancement process. Subsequently, the dual-branch structure performed enhancement processing simultaneously, where the multi-head self-attention (MHSA) mechanism was capable of capturing the long-range dependence and the discrete wavelet transform (DWT) module was employed to expand the receptive field of the proposed model. Additionally, the self-calibrated feature fusion (SCFF) module was introduced, which adaptively fused the output of each branch based on variance modulation. The evaluation experiments on LOL-V2 dataset and real samples of airport nighttime images demonstrate that, compared to EnlightenGAN, RUAS, and PairLIE method, the proposed method provides superior dynamic range and processing speed for airport real-time surveillance.