Study on n-γ discrimination based on LabVIEW KNN

With the widespread application of machine learning in neutron-gamma (n-γ) scenarios, assessing the performance and applications of machine learning in real-time discrimination scenarios has significant practical importance. LabVIEW enables tight integration of data with hardware, facilitating rapid prototyping and customizable development. The k-nearest neighbor (KNN) algorithm is a fundamental and straightforward technique in machine learning.

Optimizing KNN to achieve enhanced accuracy and superior real-time performance in n-γ discrimination, developing a virtual instrument platform to provide researchers with a more convenient method for studying real-time n-γ discrimination algorithms without the need for physical experiments.

Firstly, this study uses a Gaussian mixture model (GMM) to obtain the training set. Then, we implement the K-nearest neighbor (KNN) algorithm in the LabVIEW programming environment. Finally, the model is deployed on a development computer with a field-programmable gate array (FPGA). By comparing the classification accuracy of GMM-KNN with that of charge comparison method (CCM) in the feature space, we effectively address the issue of inaccurate labeling for pulses in the low-energy domain.

Experimental results demonstrate that KNN correctly classifies 5.52% of gamma rays, showing higher accuracy than CCM, with an average pulse processing time of only 67 μs, indicating substantial real-time performance. Using a development computer equipped with a FPGA terminal, KNN not only achieves higher precision real-time n-γ discrimination but also provides a virtual instrument platform for n-γ discrimination.