Study on neutron-gamma discrimination methods based on GMM-KNN and LabVIEW implementation

摘要: Machine learning algorithms are considered to be effective methods to improve the effectiveness of neutron-gamma (n-γ) discrimination. In this paper, we propose an intelligent discrimination method, GMM-KNN, which combines Gaussian Mixture Model with K Nearest Neighbor. Firstly, the unlabeled training and test data is categorized into 3 energy ranges: 0-25 keV, 25-100 keV, and 100-2100 keV. Secondly, GMM-KNN achieves small batch clustering in three energy intervals with only the tail integral Qtail and the total integral Qtotal as the pulse features. Then we select pulses with probability greater than 99% from the GMM clustering results to construct training set. Finally, we improve the KNN algorithm, so that GMM-KNN achieves the classification and regression algorithms through LabVIEW language, the outputs of GMM-KNN are the category or regression predictions. The GMM-KNN not only constructs the training set using unlabeled real pulse data, but also achieves n-γ discrimination of 241Am-Be pulses with the LabVIEW program. The experimental results show that GMM-KNN is highly robust and flexible. Even when using only 1/4 of the training set, the execution time of GMM-KNN is only 2021ms, with a difference of just 0.13% compared to the results obtained with the full training set. Furthermore, the accuracy of GMM-KNN is superior to that of the Charge Comparison Method (CCM), correctly classifying 5.52% of the ambiguous pulses. In addition, the GMM-KNN regressor achieves a higher Figure of Merit (FOM), with FOM values of 0.877, 1.262 and 1.020 corresponding to the three energy ranges, in particular showing a 32.08% improvement in the 0-25 keV. In conclusion, the GMM-KNN algorithm demonstrates accurate and readily deployable real-time n-γ discrimination performance, rendering it suitable for on-site analysis.