Scholarlink

<p>&#xa0; The diffusion cloud chamber is a key detector for subatomic particle research and education, yet its practical utility is severely hindered by the absence of accessible automated track analysis tools, with manual processing suffering from inefficiency, human fatigue-induced errors and inconsistent annotation. To tackle these issues, we propose a lightweight real-time automated system for cloud chamber particle track detection and analysis, based on deep learning and computer vision with a monocular camera. &#xa0; The system integrates MOG2 background subtraction, an improved YOLOv10 detector, spatial contour merging and linear fitting to realize automated identification, classification and quantitative analysis of electron, muon and alpha particle tracks in cluttered cloud chamber environments. To address the core challenges of low track-background contrast, high stochastic noise and non-physical spurious detections in cloud chamber imagery, we innovatively embed wavelet transform convolution (WTConv) and physics-informed loss (PIL) into the model framework: WTConv replaces the first convolutional layer of YOLOv10 to leverage multi-scale decomposition for suppressing background noise and enhancing high-frequency edge features of faint, slender particle tracks, while PIL enforces physical consistency with Lorentz force dynamics to penalize non-physical track predictions, thus reducing false positives and improving classification robustness. &#xa0; The synergistic effect of WTConv and PIL boosts detection precision from 94.4 to 95.2&#xa0; The proposed system resolves the long-standing bottleneck of manual track interpretation in traditional cloud chamber experiments, delivering an automated, low-cost and easily deployable solution for quantitative particle track analysis. It greatly facilitates the practical transformation of particle physics teaching, and the framework exhibits strong generalization for different cloud chamber devices, which can be adapted with minimal modification to bubble chambers and similar track detectors, providing a new paradigm for the intelligent development of nuclear physics educational tools.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

WTConv-enhanced YOLOv10 with physics-informed loss for automated particle track recognition in cloud chambers

  • 期刊论文

  • 发表日期：

    2026年5月5日

查看全文
  • Zhang Shaoru,
  • Ma Rui,
  • Liu Jia

摘要

  The diffusion cloud chamber is a key detector for subatomic particle research and education, yet its practical utility is severely hindered by the absence of accessible automated track analysis tools, with manual processing suffering from inefficiency, human fatigue-induced errors and inconsistent annotation. To tackle these issues, we propose a lightweight real-time automated system for cloud chamber particle track detection and analysis, based on deep learning and computer vision with a monocular camera.   The system integrates MOG2 background subtraction, an improved YOLOv10 detector, spatial contour merging and linear fitting to realize automated identification, classification and quantitative analysis of electron, muon and alpha particle tracks in cluttered cloud chamber environments. To address the core challenges of low track-background contrast, high stochastic noise and non-physical spurious detections in cloud chamber imagery, we innovatively embed wavelet transform convolution (WTConv) and physics-informed loss (PIL) into the model framework: WTConv replaces the first convolutional layer of YOLOv10 to leverage multi-scale decomposition for suppressing background noise and enhancing high-frequency edge features of faint, slender particle tracks, while PIL enforces physical consistency with Lorentz force dynamics to penalize non-physical track predictions, thus reducing false positives and improving classification robustness.   The synergistic effect of WTConv and PIL boosts detection precision from 94.4 to 95.2  The proposed system resolves the long-standing bottleneck of manual track interpretation in traditional cloud chamber experiments, delivering an automated, low-cost and easily deployable solution for quantitative particle track analysis. It greatly facilitates the practical transformation of particle physics teaching, and the framework exhibits strong generalization for different cloud chamber devices, which can be adapted with minimal modification to bubble chambers and similar track detectors, providing a new paradigm for the intelligent development of nuclear physics educational tools.