Thermal turbulence mitigation in underwater images of nuclear fuel assemblies

Abstract

Underwater imaging of nuclear fuel assemblies is crucial for inspection, monitoring, and safety evaluation in nuclear facilities. However, thermal turbulence caused by temperature gradients and convective flows in the cooling water can lead to severe visual degradation, including geometric distortions and blurring. To facilitate research in this underexplored area, we construct a dedicated dataset that captures thermal turbulence in underwater nuclear fuel assembly imaging. The dataset contains multi-frame sequences of turbulence-degraded images, along with corresponding ground truth images captured under still-water conditions. Building upon this dataset, we propose a novel multi-frame turbulence removal network that exploits temporal redundancy and motion cues for effective restoration. The proposed architecture integrates five key components: a feature extraction backbone for spatial encoding, a temporal self-attention block to capture long-range inter-frame dependencies, a bidirectional flow-guided propagation module, an optical flow-based warping mechanism for spatial alignment, and a fusion-reconstruction head for generating high-quality reference frames. Extensive experiments on the proposed dataset demonstrate that our method achieves superior performance over existing baselines, particularly in scenarios involving complex turbulence dynamics and non-rigid motion. The proposed framework provides a robust solution for visual enhancement in thermally dynamic underwater environments encountered in nuclear engineering applications.