Automated Neutron Dosimetry with Bubble Detectors Using CNN-Based Object Detection

摘要: Bubble detectors are widely used for measuring neutron yields in nuclear fusion and various neutron sources, as well as in monitoring personal neutron radiation dose. The neutron yield or radiation dose is typically quantified by counting the bubbles formed within the detector. However, manually counting a high density of bubbles in high-neutron-yield experiments is not only labor-intensive and time-consuming but also prone to subjective bias and errors. Futhermore, conventional threshold-based image processing algorithms face challenges in resolving issues related to bubble occlusion and overlapping, limiting their accuracy in high-density bubble scenarios. To address these challenges, an automated neutron dosimetry system that employs Convolutional Neural Network (CNN)-based object detection and a multi-view tracking algorithm has been developed. This system automates bubble quantification while minimizing counting discrepancies arising from bubble occlusion and overlapping in three-dimensional volumes. Notably, the automated counts achieved relative accuracies exceeding 95% when compared to expert manual counts on the independent test set. By capturing multi-angle imagery and performing cross-view tracking, the system effectively identifies bubbles occluded in single-perspective views. This automated framework significantly improves the precision, efficiency, and reliability of neutron dosimetry across diverse applications, including nuclear fusion, various neutron sources, and radiation protection, etc.