Neutron-gamma discrimination based on STFT-DFF model and FPGA implementation.

Area of Science:

• Nuclear Physics
• Signal Processing
• Machine Learning

Background:

• Traditional neutron-gamma discrimination methods struggle with low-energy detection.
• Existing machine learning algorithms face challenges in hardware deployment and real-time application.

Purpose of the Study:

• To develop a real-time neutron-gamma discrimination algorithm suitable for FPGA deployment.
• To improve discrimination performance, especially in the low-energy region, and enhance real-time processing efficiency.

Main Methods:

• Implemented a novel machine learning algorithm, STFT-DFF, integrating Short-Time Fourier Transform (STFT) and Dynamic Feature Fusion (DFF).
• Utilized STFT to convert pulse signals into time-frequency feature vectors, extracting neutron and gamma-ray differences.
• Employed a lightweight DFF model with an attention mechanism for adaptive feature enhancement and noise suppression.

Main Results:

• The STFT-DFF algorithm demonstrated significantly improved Figure of Merit (FOM) compared to the Charge Comparison Method (CCM) on two distinct datasets.
• Achieved real-time pulse discrimination processing efficiency exceeding 20,000 pulses per second after FPGA deployment.
• Successfully extracted time-frequency feature differences, enhancing discrimination accuracy.

Conclusions:

• The proposed STFT-DFF algorithm offers a robust solution for real-time neutron-gamma discrimination.
• FPGA deployment enables high-speed processing, making the algorithm practical for various applications.
• The integration of STFT and DFF effectively addresses the limitations of previous methods.