Low-cost CR-39 detectoralpha track automatic counting method based on YOLOv8 object detection model

Abstract: This study proposes a low-cost automated method for alpha track recognition and counting on CR-39 detec tors, providing an effective tool for radon measurement based on CR-39 detectors. By integrating a microcon troller platform with the PY-QT library, a CR-39 image acquisition system was developed, capable of capturing a complete CR-39 detector into 527 images. An automatic recognition and counting model based on YOLOv8m was trained using a dataset of over 80000 alpha track samples. The model is capable of automatically identify ing and counting alpha tracks in images. To evaluate the model’s performance, 16 etched CR-39 detectors with varying alpha track densities were tested. These detectors were sourced from granite-related experiments or high radon concentration experiments, with alpha track counts ranging from several thousand to tens of thousands. The test results show that the percentage error of the counts provided by the YOLOv8m model remains within an ideal range, typically below 1%. Meanwhile, the track sizes on these detectors are primarily concentrated in the ranges of 12-22 µm (width) × 12-22 µm (length) and 30-40 µm (width) × 30-40 µm (length). Addition ally, the study identifies the main factors affecting counting accuracy and proposes corresponding optimization solutions