Trapezoidal pile-up nuclear pulse parameter identification method based on deep learning transformer model.

Area of Science:

• Nuclear physics
• Signal processing
• Machine learning

Background:

• Pile-up of adjacent nuclear pulses is a common issue in radiation detection.
• Existing deep learning methods, like recurrent neural networks (RNNs), struggle with recognizing closely spaced pile-up pulses.

Purpose of the Study:

• To develop a more accurate and efficient method for identifying pile-up nuclear pulse parameters.
• To address the limitations of traditional deep learning models in handling short-interval pile-up events.

Main Methods:

• Application of a Transformer model, utilizing an attention mechanism, for pile-up nuclear pulse recognition.
• Comparison with traditional deep learning approaches to evaluate performance.

Main Results:

• The Transformer model demonstrates superior performance in recognizing pile-up nuclear pulses, especially those with short intervals.
• Improved accuracy and efficiency in pile-up pulse identification compared to RNNs.

Conclusions:

• The Transformer model offers a promising solution for accurate pile-up nuclear pulse recognition.
• This advancement aids in effective spectrum correction, particularly in high count rate scenarios.