Semantic segmentation with a sparse convolutional neural network for event reconstruction in MicroBooNE
P. Abratenko, M. Alrashed, R. An, J. Anthony, J. Asaadi, A. Ashkenazi, S. Balasubramanian, B. Baller, C. Barnes, G. Barr, V. Basque, L. Bathe-Peters, O. Benevides Rodrigues, S. Berkman, A. Bhanderi, A. Bhat, M. Bishai, A. Blake, T. Bolton, L. Camilleri, D. Caratelli, I. Caro Terrazas, R. Castillo Fernández, F. Cavanna, G. B. Cerati, Y. Chen, E. Church, D. Cianci, J. M. Conrad, M. Convery, L. Cooper-Troendle, J. I. Crespo-Anadón, M. Del Tutto, S. R. Dennis, Ann Devitt, Z. Djurcic, R. Dorrill, K. Duffy, S. Dytman, B. Eberly, A. Ereditato, J. J. Evans, G. A. Fiorentini Aguirre, R. S. Fitzpatrick, B. T. Fleming, N. Foppiani, D. Franco, A. P. Furmanski, D. García-Gámez, S. Gardiner, G. Ge, S. Gollapinni, O. Goodwin, E. Gramellini, P. Green, H. Greenlee, Wei Gu, R. Guénette, P. Guzowski, L. Hagaman, E. Hall, P. Hamilton, O. Hen, G. A. Horton-Smith, A. Hourlier, R. Itay, C. James, J. Jan de Vries, X. Ji, L. Jiang, J. H. Jo, R. A. Johnson, Y.-J. Jwa, N. Kamp, N. Kaneshige, G. Karagiorgi, W. Ketchum, B. Kirby, M. Kirby, T. Kobilarcik, I. Kreslo, R. LaZur, I. Lepetic, Kunxi Li, Yuzhu Li, B. R. Littlejohn, W. C. Louis, X. Luo, A. Marchionni, C. Mariani, D. Marsden, J. Marshall, J. Martín-Albo, D. A. Martínez Caicedo, K. Mason, A. Mastbaum, N. McConkey, V. Meddage, T. Mettler, K. Miller
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 MicroBooNEs ${\ensuremath{\nu}}_{e}$-appearance oscillation analyses. The network is trained to categorize pixels into five classes, which are reclassified 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 $\ensuremath{\ge}99%$. For full neutrino interaction simulations, the time for processing one image is $\ensuremath{\approx}0.5\text{ }\text{ }\mathrm{sec}$, the memory usage is at 1 GB level, which allows utilization of most typical CPU worker machine.