Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Rings of Light, Speed of AI: YOLO for Cherenkov Reconstruction

Martino Borsato, Giovanni Laganà, Maurizio Martinelli

TL;DR

The paper tackles the computational bottleneck of Cherenkov ring reconstruction for particle identification in RICH detectors by introducing a YOLO-based per-ring classifier. It uses momentum-bin closed-set classification with polar-transform ring images and annular masking to achieve real-time inference. The approach yields kaon identification efficiencies above 95% with pion misidentification around 1% across a broad momentum range and ~20 ms per full event on a high-end GPU, though a bin-dispatch overhead is currently a bottleneck. Compared with a maximum-likelihood fitter, YOLO is competitive in the momentum region most relevant for RICH1 and offers strong potential for online trigger applications, with plans to unify the model to remove dispatch overhead.

Abstract

Cherenkov rings play a crucial role in identifying charged particles in high-energy physics (HEP) experiments. Most Cherenkov ring pattern reconstruction algorithms currently used in HEP experiments rely on a likelihood fit to the photo-detector response, which often consumes a significant portion of the computing budget for event reconstruction. We present a novel approach to Cherenkov ring reconstruction using YOLO, a computer vision algorithm capable of real-time object identification with a single pass through a neural network. We obtain a reconstruction efficiency above 95% and a pion misidentification rate below 5% across a wide momentum range for all particle species.

Rings of Light, Speed of AI: YOLO for Cherenkov Reconstruction

TL;DR

The paper tackles the computational bottleneck of Cherenkov ring reconstruction for particle identification in RICH detectors by introducing a YOLO-based per-ring classifier. It uses momentum-bin closed-set classification with polar-transform ring images and annular masking to achieve real-time inference. The approach yields kaon identification efficiencies above 95% with pion misidentification around 1% across a broad momentum range and ~20 ms per full event on a high-end GPU, though a bin-dispatch overhead is currently a bottleneck. Compared with a maximum-likelihood fitter, YOLO is competitive in the momentum region most relevant for RICH1 and offers strong potential for online trigger applications, with plans to unify the model to remove dispatch overhead.

Abstract

Cherenkov rings play a crucial role in identifying charged particles in high-energy physics (HEP) experiments. Most Cherenkov ring pattern reconstruction algorithms currently used in HEP experiments rely on a likelihood fit to the photo-detector response, which often consumes a significant portion of the computing budget for event reconstruction. We present a novel approach to Cherenkov ring reconstruction using YOLO, a computer vision algorithm capable of real-time object identification with a single pass through a neural network. We obtain a reconstruction efficiency above 95% and a pion misidentification rate below 5% across a wide momentum range for all particle species.

Paper Structure

This paper contains 11 sections, 2 equations, 2 figures.

Table of Contents

  1. Introduction
  2. Dataset
  3. YOLO Classification Pipeline
  4. Training and evaluation
  5. Experiments
  6. Grouping vs non-grouping.
  7. Model size.
  8. Image size.
  9. Optimal pipeline.
  10. Comparison with Maximum-Likelihood fit
  11. Conclusions

Figures (2)

  • Figure 1: From full RICH1 event to masked polar view. Left: full event (selected ring in red); middle: single-ring image in Cartesian coordinates with highlighted hits; right: annular mask and polar transform $(r,\theta)$. In middle and right, all lit pixels are set to 255 for clarity.
  • Figure 2: Comparison of YOLO Object Detection Performance and Likelihood Estimation.