The Automatic Calibration Method of the Compton Edge Based on Normalized Cross-correlation and Simulated Annealing Algorithm

TL;DR

This paper introduces an automated calibration method for Compton edges in scintillation detectors using normalized cross-correlation, simulated annealing, and a convolutional response model, improving accuracy and stability over traditional techniques.

Contribution

It presents a novel automated calibration approach combining NCC, SA, and convolutional modeling, addressing limitations of existing methods in low count and spectral overlap scenarios.

Findings

Achieves calibration accuracy comparable to full-energy peak methods (cosine similarity >99.999%)

Demonstrates superior stability over traditional methods in various spectral conditions

Reduces average errors significantly in low count and spectral overlap cases

Abstract

Accurate energy channel calibration in scintillation detectors is essential for reliable radiation detection across nuclear physics, medical imaging, and environmental monitoring. Organic scintillators like BC408 and EJ309 lack full-energy peaks, making their Compton edge a critical calibration alternative where traditional peak methods fail. Existing Compton edge identification techniques - Gaussian fitting for the 50%-70% amplitude point, first derivative minimum detection, and Monte Carlo simulation - suffer significant degradation from low count rates, spectral overlap, and subjective interval selection. For the first time, we propose an automated calibration procedure based on Normalized Cross-Correlation (NCC), Simulated Annealing (SA), and a convolutional response model to address these issues. This method automates the selection of the Compton edge interval through NCC-based matching, utilizes SA for global parameter optimization, and then employs a convolutional model for precise matching. Experiments involving the irradiation of organic scintillators (BC408, EJ309) and inorganic scintillators (NaI:Tl, LaBr3:Ce) with 137Cs, 22Na, 54Mn, and 60Co radiation sources demonstrate that this method achieves accuracy commensurate with full-energy peak calibration method (cosine similarity >99.999%) and exhibits superior stability compared to the two traditional methods. In the extreme cases of spectral overlap and low count rate, the average errors of this method are 19.77% and 15.65% of those from the two traditional methods in BC408, 56.44% and 33.15% of those from the two traditional methods in EJ309. This work advances detector calibration and offers a scalable, automated solution for high-energy experiments and portable devices.