Speaker
Description
The study of nuclear structure, particularly clustering phenomena in light nuclei such as Carbon-12 (¹²C), is essential for advancing our understanding of nuclear forces and stellar nucleosynthesis. Elastic and inelastic scattering reactions, like ¹²C(p,p)¹²C and ¹²C(α,α)¹²C, are powerful probes for investigating these fundatmental properties. The Active Target Time Projection Chamber (AT-TPC), such as the MATE-TPC developed at Institute of Modern Physics, offers comprehensive 3D tracking of reaction products, allowing for precise reconstruction of reaction kinematics and missing-mass spectra. However, the large volume and complexity of the resulting point-cloud data pose significant challenges for conventional data analysis methods, which are often labor-intensive and time-consuming.
To overcome this challenges, we have developed an automated event classification framework utilizing a deep convolutional neural network (CNN). In this work, we apply the ResNet-18 architecture to classify reaction events from ¹²C+p and ¹²C+α scatterings simulated with the MATESIM toolkit in the MATEROOT code. The 3D particle tracks “detected” by the MATE-TPC are projected onto 2D planes, generating image-like input for the network. Our trained model achieves high performance, attaining over 94% classification accuracy in distinguishing proton events from none-proton events. Analysis of the model's performance, including its confusion matrix, and examination of misclassified events demonstrate its robustness and identifies areas for further improvement. This deep learning approach greatly enhances the efficiency and objectivity of data analysis in AT-TPC experiments, paving the way for rapid processing of large datasets and enabling searches for rare reaction channels that are essential for advancing nuclear physics research.
| Research field of your presentation | Experimental Low-energy nuclear physics |
|---|
