Research on Image Denoising Algorithm in Nuclear Accident Rescue Scenarios

In search-and-rescue missions, unmanned aerial vehicles (UAVs) predominantly rely on visual sensors for information acquisition. However, in nuclear accident scenarios, high-energy particles severely degrade imaging sensors, significantly compromising visual perception in radioactive environments. This study investigates the characteristic patterns of radiation-induced image noise and proposes a deep learning-based denoising algorithm to enhance video quality. The proposed method employs localized convolutions to capture fine-grained noise textures while integrating multi-head attention mechanisms for global contextual modeling. Experimental results demonstrate performance improvements of 1.2% in PSNR and 3.1% in SSIM metrics over baseline methods, validating the technical efficacy of our approach.