Semantic Segmentation with a Sparse Convolutional Neural Network for Event Reconstruction in MicroBooNE

Physics

Associated organisational unit

Experimental Particle Physics

Electronic data

Sparse_SSNet_Paper_v7
Rights statement: © 2021 American Physical Society
Accepted author manuscript, 2.25 MB, PDF document
Available under license: CC BY-NC: Creative Commons Attribution-NonCommercial 4.0 International License

Text available via DOI:

https://doi.org/10.1103/PhysRevD.103.052012
Final published version

Keywords

physics.ins-det, hep-ex

View graph of relations

Research output: Contribution to Journal/Magazine › Journal article › peer-review

Published

Standard

Semantic Segmentation with a Sparse Convolutional Neural Network for Event Reconstruction in MicroBooNE. / MicroBooNE Collaboration ; Blake, A.; Devitt, D. et al.
In: Physical Review D, Vol. 103, No. 5, 052012, 26.03.2021.

Research output: Contribution to Journal/Magazine › Journal article › peer-review

Bibtex

@article{7796c76454db4cd48978fc9e5286a1a6,

title = "Semantic Segmentation with a Sparse Convolutional Neural Network for Event Reconstruction in MicroBooNE",

abstract = " We present the performance of a semantic segmentation network, SparseSSNet, that provides pixel-level classification of MicroBooNE data. The MicroBooNE experiment employs a liquid argon time projection chamber for the study of neutrino properties and interactions. SparseSSNet is a submanifold sparse convolutional neural network, which provides the initial machine learning based algorithm utilized in one of MicroBooNE's $\nu_e$-appearance oscillation analyses. The network is trained to categorize pixels into five classes, which are re-classified into two classes more relevant to the current analysis. The output of SparseSSNet is a key input in further analysis steps. This technique, used for the first time in liquid argon time projection chambers data and is an improvement compared to a previously used convolutional neural network, both in accuracy and computing resource utilization. The accuracy achieved on the test sample is $\geq 99\%$. For full neutrino interaction simulations, the time for processing one image is $\approx$ 0.5 sec, the memory usage is at 1 GB level, which allows utilization of most typical CPU worker machine. ",

keywords = "physics.ins-det, hep-ex",

author = "{MicroBooNE Collaboration} and MicroBooNE collaboration and P. Abratenko and M. Alrashed and R. An and J. Anthony and J. Asaadi and A. Ashkenazi and S. Balasubramanian and B. Baller and G. Barr and V. Basque and L. Bathe-Peters and Rodrigues, {O. Benevides} and S. Berkman and A. Bhanderi and A. Bhat and M. Bishai and A. Blake and T. Bolton and L. Camilleri and D. Caratelli and Terrazas, {I. Caro} and Fernandez, {R. Castillo} and F. Cavanna and G. Cerati and E. Church and D. Cianci and Conrad, {J. M.} and M. Convery and L. Cooper-Troendle and Crespo-Anadon, {J. I.} and Tutto, {M. Del} and Dennis, {S. R.} and D. Devitt and R. Diurba and R. Dorrill and K. Duffy and S. Dytman and B. Eberly and A. Ereditato and Aguirre, {G. A. Fiorentini} and Fitzpatrick, {R. S.} and Fleming, {B. T.} and N. Foppiani and D. Franco and Furmanski, {A. P.} and D. Garcia-Gamez and S. Gardiner and J. Nowak and C. Thorpe",

note = "{\textcopyright} 2021 American Physical Society ",

year = "2021",

month = mar,

day = "26",

doi = "10.1103/PhysRevD.103.052012",

language = "English",

volume = "103",

journal = "Physical Review D",

issn = "1550-7998",

publisher = "American Physical Society",

number = "5",

}

RIS

TY - JOUR

T1 - Semantic Segmentation with a Sparse Convolutional Neural Network for Event Reconstruction in MicroBooNE

AU - MicroBooNE Collaboration

AU - collaboration, MicroBooNE

AU - Abratenko, P.

AU - Alrashed, M.

AU - An, R.

AU - Anthony, J.

AU - Asaadi, J.

AU - Ashkenazi, A.

AU - Balasubramanian, S.

AU - Baller, B.

AU - Barr, G.

AU - Basque, V.

AU - Bathe-Peters, L.

AU - Rodrigues, O. Benevides

AU - Berkman, S.

AU - Bhanderi, A.

AU - Bhat, A.

AU - Bishai, M.

AU - Blake, A.

AU - Bolton, T.

AU - Camilleri, L.

AU - Caratelli, D.

AU - Terrazas, I. Caro

AU - Fernandez, R. Castillo

AU - Cavanna, F.

AU - Cerati, G.

AU - Church, E.

AU - Cianci, D.

AU - Conrad, J. M.

AU - Convery, M.

AU - Cooper-Troendle, L.

AU - Crespo-Anadon, J. I.

AU - Tutto, M. Del

AU - Dennis, S. R.

AU - Devitt, D.

AU - Diurba, R.

AU - Dorrill, R.

AU - Duffy, K.

AU - Dytman, S.

AU - Eberly, B.

AU - Ereditato, A.

AU - Aguirre, G. A. Fiorentini

AU - Fitzpatrick, R. S.

AU - Fleming, B. T.

AU - Foppiani, N.

AU - Franco, D.

AU - Furmanski, A. P.

AU - Garcia-Gamez, D.

AU - Gardiner, S.

AU - Nowak, J.

AU - Thorpe, C.

PY - 2021/3/26

Y1 - 2021/3/26

N2 - We present the performance of a semantic segmentation network, SparseSSNet, that provides pixel-level classification of MicroBooNE data. The MicroBooNE experiment employs a liquid argon time projection chamber for the study of neutrino properties and interactions. SparseSSNet is a submanifold sparse convolutional neural network, which provides the initial machine learning based algorithm utilized in one of MicroBooNE's $\nu_e$-appearance oscillation analyses. The network is trained to categorize pixels into five classes, which are re-classified into two classes more relevant to the current analysis. The output of SparseSSNet is a key input in further analysis steps. This technique, used for the first time in liquid argon time projection chambers data and is an improvement compared to a previously used convolutional neural network, both in accuracy and computing resource utilization. The accuracy achieved on the test sample is $\geq 99\%$. For full neutrino interaction simulations, the time for processing one image is $\approx$ 0.5 sec, the memory usage is at 1 GB level, which allows utilization of most typical CPU worker machine.

AB - We present the performance of a semantic segmentation network, SparseSSNet, that provides pixel-level classification of MicroBooNE data. The MicroBooNE experiment employs a liquid argon time projection chamber for the study of neutrino properties and interactions. SparseSSNet is a submanifold sparse convolutional neural network, which provides the initial machine learning based algorithm utilized in one of MicroBooNE's $\nu_e$-appearance oscillation analyses. The network is trained to categorize pixels into five classes, which are re-classified into two classes more relevant to the current analysis. The output of SparseSSNet is a key input in further analysis steps. This technique, used for the first time in liquid argon time projection chambers data and is an improvement compared to a previously used convolutional neural network, both in accuracy and computing resource utilization. The accuracy achieved on the test sample is $\geq 99\%$. For full neutrino interaction simulations, the time for processing one image is $\approx$ 0.5 sec, the memory usage is at 1 GB level, which allows utilization of most typical CPU worker machine.

KW - physics.ins-det

KW - hep-ex

U2 - 10.1103/PhysRevD.103.052012

DO - 10.1103/PhysRevD.103.052012

M3 - Journal article

VL - 103

JO - Physical Review D

JF - Physical Review D

SN - 1550-7998

IS - 5

M1 - 052012

ER -

Research

Associated organisational unit

Electronic data

Links

Text available via DOI:

Keywords