Transforming a rare event search into a not-so-rare event search in real-time with deep learning-based object detection

J. Schueler,H. M. Araújo,S. N. Balashov,J. E. Borg,C. Brew,F. M. Brunbauer,C. Cazzaniga,A. Cottle,C. D. Frost,F. Garcia,D. Hunt,A. C. Kaboth,M. Kastriotou,I. Katsioulas,A. Khazov,P. Knights,H. Kraus,V. A. Kudryavtsev,S. Lilley,A. Lindote,M. Lisowska,D. Loomba,M. I. Lopes,E. Lopez Asamar,P. Luna Dapica,P. A. Majewski,T. Marley,C. McCabe,L. Millins,A. F. Mills,M. Nakhostin,R. Nandakumar,T. Neep,F. Neves,K. Nikolopoulos,E. Oliveri,L. Ropelewski,V. N. Solovov,T. J. Sumner,J. Tarrant,E. Tilly,R. Turnley,R. Veenhof

2024-06-12

Abstract:Deep learning-based object detection algorithms enable the simultaneous classification and localization of any number of objects in image data. Many of these algorithms are capable of operating in real-time on high resolution images, attributing to their widespread usage across many fields. We present an end-to-end object detection pipeline designed for real-time rare event searches for the Migdal effect, using high-resolution image data from a state-of-the-art scientific CMOS camera in the MIGDAL experiment. The Migdal effect in nuclear scattering, crucial for sub-GeV dark matter searches, has yet to be experimentally confirmed, making its detection a primary goal of the MIGDAL experiment. Our pipeline employs the YOLOv8 object detection algorithm and is trained on real data to enhance the detection efficiency of nuclear and electronic recoils, particularly those exhibiting overlapping tracks that are indicative of the Migdal effect. When deployed online on the MIGDAL readout PC, we demonstrate our pipeline to process and perform the rare event search on 2D image data faster than the peak 120 frame per second acquisition rate of the CMOS camera. Applying these same steps offline, we demonstrate that we can reduce a sample of 20 million camera frames to around 1000 frames while maintaining nearly all signal that YOLOv8 is able to detect, thereby transforming a rare search into a much more manageable search. Our studies highlight the potential of pipelines similar to ours significantly improving the detection capabilities of experiments requiring rapid and precise object identification in high-throughput data environments.

High Energy Physics - Experiment,Instrumentation and Detectors

What problem does this paper attempt to address?

The main objective of this paper is to demonstrate how to use the deep learning-based object detection algorithm YOLOv8 to achieve real-time rare event searches, particularly for the detection of the Migdal effect. The Migdal effect refers to the phenomenon where electrons are emitted with low probability during the scattering of an atomic nucleus, which is significant for dark matter detection in the sub-GeV mass range. However, to date, the Migdal effect has not been confirmed experimentally.

To achieve this goal, researchers developed an end-to-end object detection pipeline to process high-resolution image data from the MIGDAL experiment. The MIGDAL experiment is an experiment using a deuterium-deuterium (D-D) fusion generator to produce neutrons, aimed at detecting and measuring the Migdal effect. The imaging device used in the experiment is an advanced Complementary Metal-Oxide-Semiconductor (CMOS) camera capable of capturing images at up to 120 frames per second.

Specifically, the key contributions of the paper include:

1. **Development of the YOLOv8 pipeline**: This pipeline uses the YOLOv8 object detection algorithm and is trained on real data to improve the detection efficiency of nuclear recoils (NR) and ionization recoils (ER), especially those events exhibiting overlapping tracks, which are considered characteristic of the Migdal effect.

2. **Real-time processing capability**: The pipeline can process image data online at a peak acquisition rate of 120 frames per second, enabling rare event searches.

3. **Large-scale data screening**: By applying the same steps offline, 20 million frames of image data can be reduced to approximately 1,000 frames while retaining almost all signals detectable by YOLOv8, thus transforming the rare event search into a more manageable task.

4. **Performance evaluation**: The paper also demonstrates the performance of YOLOv8 in the identification and localization of single and multiple tracks, including the identification of ER and NR at different energy levels, and the relationship between Migdal effect detection efficiency and ER energy, NR energy, and the geometric overlap of ER and NR.

In summary, this study not only provides new tools and techniques for the direct detection of the Migdal effect but also demonstrates the potential application value of deep learning-based object detection algorithms in high-throughput data environments requiring rapid and accurate object identification.