Image restoration of underwater fuel assembly thermal turbulence based on improved U-Net network

As a huge heat source, the fuel assembly directly causes its surrounding dynamic temperature field and subsequently induces the irregular real-time variations of refractive index. Consequently, thermal turbulence phenomena inevitably occur, affecting the measurement accuracy when measuring the bow and twist deformation of fuel assembly. To mitigate the impact brought by the thermal turbulence, an improved U-Net network that boasts exceptional restoration efficiency and precision is presented. Utilizing a linear activation function, corresponding experimental datasets are captured and preprocessed for further training. The principle of thermal turbulence phenomenon is elaborated and the corresponding existence is demonstrated by field turbulence experiments. Compared to thermal turbulence conditions, field restoration experiments demonstrate that our restoration model significantly reduces the impact of image fluctuations and temporal fluctuations, prominently enhances the accuracy of center coordinate extraction, and subsequently provides a strong guarantee for the field deformation measurement of fuel assembly.