Vision Calorimeter for High-Energy Particle Detection

TL;DR

ViC reframes anti-neutron parameter estimation as an object-detection problem on EMC-derived particle images, leveraging a physics-informed heat-conduction representation to extract robust features. It introduces HCO-K with 2-D DCT/IDCT operations and HeatK layers to bridge natural and particle-image domains, and uses pseudo bounding boxes to enable end-to-end training. The momentum head exploits global attention while the position head relies on local cues, with momentum regression formulated as $p_{pred}=p_{base}\cdot e^{p_{out}}$ and $p_{base}=1.0$. On BESIII EMC data, ViC achieves $mAB=9.32^{\circ}$ and $mRE=21.48\%$, significantly outperforming clustering baselines and enabling invariant mass reconstruction, demonstrating practical impact for high-energy physics analyses.

Abstract

In high-energy physics, estimating anti-neutron parameters (position and momentum) using the electromagnetic calorimeter (EMC) is crucial but challenging. To conquer this challenge, we propose Vision Calorimeter (ViC), a framework that migrates visual object detectors to analyze particle images. The motivation lies in introducing a physics-inspired heat-conduction operator (HCO) into the detector's backbone and head to handle the discrete and sparse patterns of these images. Implemented via the Discrete Cosine Transform, HCO extracts frequency-domain features, bridging the distribution gap between natural and particle images. Experiments demonstrate that ViC significantly outperforms conventional methods, reducing the incident position prediction error by 46.16% (from 17.31° to 9.32°) and providing the first baseline result with an incident momentum regression error of 21.48%. This study underscores ViC's great potential as a reliable particle detector for high-energy physics. Code is available at https://github.com/yuhongtian17/ViC.

Figures (16)

• Figure 1: Overview of the Vision Calorimeter (ViC) pipeline. ViC integrates a novel heat-conduction operator into both the backbone and head of a visual object detector. Inspired by the principles of physical heat-conduction, this operator combines a radial prior with global attention, effectively capturing the unique characteristics of particle patterns. Implemented via discrete cosine transform (DCT), it ensures seamless alignment with pre-trained visual representation, enabling efficient transfer learning from natural images to particle images. (Best viewed in color)
• Figure 2: Formatting a high-energy particle image. (a) EMC cell arrays are arranged in a spherical coordinate system to capture the spatial structure of particle interactions. (b) The energy depositions are projected to a 2-D image domain, representing the readout as pixel intensities for subsequent analysis.
• Figure 3: Histogram of deposited energy recorded by EMC.
• Figure 4: The average particle image manifests the energy deposition pattern of $\bar{n}$. After extraction by the heat-conduction operator, the features of particle images become akin to those of natural images. (The horse image shown here is sourced from the CIFAR-10 dataset cifar.)
• Figure 5: Heat-conduction operator (HCO), which integrates blocks of layer normalization (LN), feed-forward network (FFN), and depth-wise convolution (DWConv). Blocks in blue represent features of the original HCO design, while elements in red highlight the modifications introduced in HCO-K.