Study on n/γ Discrimination Method Based on Deep Metric Learning

This study proposes a deep metric learning based n/γ discrimination method. Based on an improved convolutional neural network structure and combined with triplet loss function constraints, the discriminative feature space of neutron and gamma nuclear pulses is obtained, achieving efficient discrimination of neutron and gamma rays. A mixed dataset of hardware simulated nuclear pulse generator data and software simulated data was used for training, and quantitative tests were conducted on simulated data and measured pulse data with electronic waveform time characteristic jitter. This dataset mainly includes three types of content: 1. Data generated by the hardware DT5800 simulated nuclear pulse generator; 2. Use Python software to simulate the pulse shape characteristics in actual nuclear physics experiments, and generate data using a double exponential function for simulation; 3. All figures and tables in the paper are designed with data.