Visual inspection for nuclear fuel assemblies is necessary during outages of nuclear power plants. These inspections can be used to identify fuel assemblies’ anomalies that endanger the reactor’s running. However, intense radiation of fuel assembly sensitively degrades the image quality through a mixture of impulse and Gaussian noise. To solve this problem, an image denoising algorithm based on Non-Local Dual Denoising (NLDD) and Rank-Ordered Absolute Differences (ROAD) is proposed here. It consists of two steps. The detector ROAD is first used to find noisy pixels in an image damaged by impulse noise and replace them with neighborhood values. Then, NLDD filter is applied to image corrupted with Gaussian noise and retains the details. The proposed approach has been successfully tested on assembly inspection of nuclear power plants. The results reveal that our approach is effective to noise suppression and crucial detail preservation.
CMOS image sensors rival CCDs in domains that include strong radiation resistance as well as simple drive signals, so it is widely applied in the high-energy radiation environment, such as space optical imaging application and video monitoring of nuclear power equipment. However, the silicon material of CMOS image sensors has the ionizing dose effect in the high-energy rays, and then the indicators of image sensors, such as signal noise ratio (SNR), non-uniformity (NU) and bad point (BP) are degraded because of the radiation. The radiation environment of test experiments was generated by the <sup>60</sup>Co γ-rays source. The camera module based on image sensor CMV2000 from CMOSIS Inc. was chosen as the research object. The ray dose used for the experiments was with a dose rate of 20krad/h. In the test experiences, the output signals of the pixels of image sensor were measured on the different total dose. The results of data analysis showed that with the accumulation of irradiation dose, SNR of image sensors decreased, NU of sensors was enhanced, and the number of BP increased. The indicators correction of image sensors was necessary, as it was the main factors to image quality. The image processing arithmetic was adopt to the data from the experiences in the work, which combined local threshold method with NU correction based on non-local means (NLM) method. The results from image processing showed that image correction can effectively inhibit the BP, improve the SNR, and reduce the NU.