Alpha track counting on CR-39 detectors using YOLOv8s: Model development and application

Precise counting and detection of alpha tracks on CR-39 detectors is a critical part of measuring indoor radon and thoron. Traditional software, such as ImageJ, often struggles with background artifacts and overlapping tracks. In this work, we developed a deep learning-based method using the YOLOv8s model to address these problems. Before training, we pre-processed the CR-39 images to enhance track visibility and reduce noise, thus allowing the model to focus on true alpha tracks. We then trained the selected model on the processed images. The results showed that the trained YOLOv8s achieved high precision, recall, F1-score, and mean average precision values of 0.892, 0.902, 0.897, 0.94, and 0.517, respectively. Moreover, YOLOv8s performed better than ImageJ in terms of track overlaps separation and artifacts rejection. We have also implemented an application integrating the trained model for fast, user-friendly, and automated alpha track counting. Overall, the findings of this study demonstrate that deep learning-based object detection can make alpha track counting on CR-39 detectors more accurate and less time-consuming.