利用捲積神經網路進行Belle II ECL之能量修正

Abstract

在Belle II 實驗中，CsI晶體構成的電磁量能儀[18]被用來偵測光子的能量。然而，能量滲出導致的損失降低了偵測器的準確度。這些能量滲出是晶體之間無反應的區域以及粒子穿透晶體等原因導致的。 探討了電磁量能儀中能量的散布，我們相信這些能量滲出能透過研究沉積能量的分佈模式來進行修正。根據這個想法，我們將沉積能量的分佈以二維圖片的方式表現出來，並且利用卷積神經網路的圖形辨識能力來進行能量修正。 在使用模擬資料進行訓練以及驗證後，我們得到了極具發展性的結果。我們希望在經過更進一步的研究和訓練後，此種修正方法也可以應用在真實資料上。

For ECL in Belle II experiment [18], CsI crystals are used to detect photon energy. However, the energy loss caused by leakage decrease the accuracy of the detector. Leakage exists because of non-sensitive region between the scintillator crystals, and the particle penetration through the crystals. By considering the energy distribution of ECL, we believe that these leakages can be corrected by studying the patterns of distribution of deposited energy. Based on this idea, we represent the distribution of deposited energy as a 2-D image, and correct the energy loss with the pattern recognition ability of convolutional neural network. We got promising result from training and validating CNN model with simulation data. With further studying and training, we hope this kind of correction can apply on real data as well.