Image reconstruction method of gamma emission tomography based on prior-aware information and machine learning for partial-defect detection of PWR-type spent nuclear fuel

Spent nuclear fuel (SNF) requires accurate and effective supervision to prevent radiation release to the public and the environment. Gamma emission tomography (GET) has been used to inspect partial-defects at the pin-by-pin level through the acquisition of tomographic images. However, due to the high density of nuclear fuel rods, GET shows relatively low accuracy for the central region. This study proposes an image reconstruction method for the internal fuels based on Monte Carlo simulation and machine learning algorithm. The gammas from the SNF were measured with the GET device and the sinograms were reconstructed with Maximum-Likelihood Expectation-Maximization (MLEM) with prior-aware information of SNF, and image quality was improved with the machine learning technique. Our results show that the MLEM with prior-aware information could dramatically improve the image quality and the denoising technique with the machine learning could clearly correct the over-and under-estimation. Based on the results of quantitative validation of the proposed reconstruction method, the pattern classification accuracy was calculated to be about 100 %. These techniques enabled the image acquisition of the central fuel rod and increased accuracy in partial defect detection. For further study, the performance of the proposed reconstruction method will be evaluated based on the experiment data.