DeepHCC: Deep learning model for real-time count rate determination of particles in homemade cloud chamber

Abstract

Abstract Cosmic ray (CR) particles are high-energy particles originating from outer space that often reach the ground as alpha and proton particles, which can be studied using particle detectors. While investigating CR particles is a complex process, cloud chambers offer a simple means of detection. However, they cannot measure count rates directly. To overcome this limitation, we developed a deep learning model for real-time count rate determination of particles in a homemade cloud chamber (DeepHCC), implementing YOLOv5 pre-trained weights to use with our homemade cloud chamber. We trained DeepHCC using a total of 2,435 images from our homemade cloud chamber and other cloud chambers and found that YOLOv5m was the most suitable model for our task due to its fast and accurate detection capabilities. DeepHCC performed well in fundamental evaluations, with an overall area under the precision-recall curve of 0.8886 and an F1 score of 0.8624 while operating at 61 frames per second. We also experimented on count rate determination compared to human performance, which yielded an overall accuracy of 80.15%. Overall, DeepHCC could be used with our homemade cloud chambers to create particle detector kits with comparable performance. However, our model could not detect muons and accurately identify electrons due to an imbalanced dataset used in the training process. These reveal opportunities for future research to explore other techniques to address this problem. We also plan to compare our results with other detectors to verify reliability. The results demonstrate that our work could also be an innovative prototype for an engaging educational physics instrument.