Neutron and gamma discrimination based on multi-features and KNN-LDA

摘要: In order to improve the n/γ discrimination effect, this paper proposes an intelligent discrimination method based on multi-features and KNN-LDA algorithm. Firstly, this paper establishes 6 feature parameters covering time and frequency domains according to the PSD principle; secondly, an automatic feature extraction system is designed, which can intelligently extract the optimal distributions of multi-features from the pulse data. Then, the feature criterion is constructed and calculated based on the feature distributions, so as to divide reliable data from the feature data as the training set of the model, and the rest as the test set. Finally, based on the training set, using the regression optimization and dimensionality reduction of the KNN-LDA model, the test set is input into the model for further classification to achieve the n/γ discrimination of all pulses. According to the experimental results, the FOM value of the KNN-LDA model reaches 3.07 in the test set of high-energy domain (≥40 keV), which is 245% higher than that of the CCM; in the test set of low-energy domain (≤40 keV), the FOM value of the KNN-LDA model reaches 2.64, which is 355% higher than that of the CCM. The experimental results show that the discrimination effect of this method is excellent, especially the discrimination of low-energy-domain is greatly improved, which provides a new idea for introducing supervised learning model to solve the difficult problem of low-energy-domain discrimination.