A methodology for alpha/beta particles identification in Liquid Scintillation using a three-channel Convolutional Neural Network

Fig. 1 Schematic diagram of the 3-PMTs detection system in a liquid scintillation spectrometerFig.2 Waveforms of fast pulses generated by liquid scintillation counter; (a) Comparison between alpha pulse and beta pulse; (b) 100 alpha pulses; (c) 100 beta pulses; (d) The fast pulses output by three photomultiplier tubes, pulse CHA from PMTA, pulse CHB from PMTB, and pulse CHC from PMTCFig.3 The architecture of the Three-channel Convolutional Neural NetworkFig.4 (a) The impact of the number of training pulses on the accuracy of the validation set; (b) The impact of epoch on the accuracy of validation setFig.5 Distribution of segmentation parameters from composite sample; (a) One-dimensional distributions produced by CI, CNN, and TCNN; (b) Two-dimensional distribution (charge ratio, height) produced by CI; (c) Two-dimensional distribution (z1, height) produced by CNN; (d) Two-dimensional distribution (z1, height) produced by TCNNFig.6 The spectrum from the composite sample; (a) The unseparated spectrum; (b) The separated spectrum generated by CI; (c) The separated spectrum generated by CNN; (d) The separated spectrum generated by TCNNFig.7 Analysis of α-MCA spectra from composite sample; (a) Precision-recall curve; (b) F1score-recall curve; (c) LOD-recall curve; (d) Background-recall curve