Saúl Alonso-Monsalve, Davide Sgalaberna, Xingyu Zhao, Adrien Molines, Clark McGrew, André Rubbia
{"title":"Deep-learning-based decomposition of overlapping-sparse images: application at the vertex of simulated neutrino interactions","authors":"Saúl Alonso-Monsalve, Davide Sgalaberna, Xingyu Zhao, Adrien Molines, Clark McGrew, André Rubbia","doi":"10.1038/s42005-024-01669-8","DOIUrl":null,"url":null,"abstract":"Image decomposition plays a crucial role in various computer vision tasks, enabling the analysis and manipulation of visual content at a fundamental level. Overlapping and sparse images pose unique challenges for decomposition algorithms due to the scarcity of meaningful information to extract components. Here, we present a solution based on deep learning to accurately extract individual objects within multi-dimensional overlapping-sparse images, with a direct application to the decomposition of overlaid elementary particles obtained from imaging detectors. Our approach allows us to identify and measure independent particles at the vertex of neutrino interactions, where one expects to observe images with indiscernible overlapping charged particles. By decomposing the image of the detector activity at the vertex through deep learning, we infer the kinematic parameters of the low-momentum particles and enhance the reconstructed energy resolution of the neutrino event. Finally, we combine our approach with a fully-differentiable generative model to improve the image decomposition further and the resolution of the measured parameters. This improvement is crucial to search for asymmetries between matter and antimatter. The paper addresses the task of extracting individual objects from multi-dimensional overlapping-sparse images, with valuable impact in high-energy physics (future high-precision long-baseline neutrino oscillation experiments). The developed tool will allow to reduce systematic errors and avoid model dependence, improving the neutrino energy resolution and sensitivity.","PeriodicalId":10540,"journal":{"name":"Communications Physics","volume":null,"pages":null},"PeriodicalIF":5.4000,"publicationDate":"2024-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.nature.com/articles/s42005-024-01669-8.pdf","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Communications Physics","FirstCategoryId":"101","ListUrlMain":"https://www.nature.com/articles/s42005-024-01669-8","RegionNum":1,"RegionCategory":"物理与天体物理","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"PHYSICS, MULTIDISCIPLINARY","Score":null,"Total":0}
引用次数: 0
Abstract
Image decomposition plays a crucial role in various computer vision tasks, enabling the analysis and manipulation of visual content at a fundamental level. Overlapping and sparse images pose unique challenges for decomposition algorithms due to the scarcity of meaningful information to extract components. Here, we present a solution based on deep learning to accurately extract individual objects within multi-dimensional overlapping-sparse images, with a direct application to the decomposition of overlaid elementary particles obtained from imaging detectors. Our approach allows us to identify and measure independent particles at the vertex of neutrino interactions, where one expects to observe images with indiscernible overlapping charged particles. By decomposing the image of the detector activity at the vertex through deep learning, we infer the kinematic parameters of the low-momentum particles and enhance the reconstructed energy resolution of the neutrino event. Finally, we combine our approach with a fully-differentiable generative model to improve the image decomposition further and the resolution of the measured parameters. This improvement is crucial to search for asymmetries between matter and antimatter. The paper addresses the task of extracting individual objects from multi-dimensional overlapping-sparse images, with valuable impact in high-energy physics (future high-precision long-baseline neutrino oscillation experiments). The developed tool will allow to reduce systematic errors and avoid model dependence, improving the neutrino energy resolution and sensitivity.
期刊介绍:
Communications Physics is an open access journal from Nature Research publishing high-quality research, reviews and commentary in all areas of the physical sciences. Research papers published by the journal represent significant advances bringing new insight to a specialized area of research in physics. We also aim to provide a community forum for issues of importance to all physicists, regardless of sub-discipline.
The scope of the journal covers all areas of experimental, applied, fundamental, and interdisciplinary physical sciences. Primary research published in Communications Physics includes novel experimental results, new techniques or computational methods that may influence the work of others in the sub-discipline. We also consider submissions from adjacent research fields where the central advance of the study is of interest to physicists, for example material sciences, physical chemistry and technologies.