Operationalizing nuclear safety: A remote α/β detection system integrating CNN and radioluminescence for emergency...

The discrimination of α/β radioactive materials during nuclear emergencies is of paramount importance due to their severe threat to public health and the environment. Conventional techniques, such as pulse shape differentiation, are limited by factors including particle recoil signal trailing, complex field interference, and short α-ray propagation, which impede rapid and precise detection. This study proposes an innovative remote α/β discrimination technique employing CNN. The proposed method involves the construction of a multilayer CNN model, which utilizes 2D α/β imaging maps from a scanning device as input. This model is capable of outputting the activity and center position coordinates of the radioactive materials. To ensure the efficacy of the model, particle transport simulations were conducted, generating 100,000 samples for training and testing purposes. Following a training phase with 80,000 data sets and a testing phase with 20,000 data sets, the model achieves a 99.1% accuracy rate, with 72.4% of position coordinate predictions exhibiting a relative error of less than 10%. Experimental validation substantiates the model's validity, thereby confirming the efficacy and reliability of the technology for radioactive substance safety monitoring in nuclear accidents, with the potential to enhance detection efficiency and mitigate hazards.