Research and application of machine learning methods in X-ray in-situ calibration

Abstract

The correction of environmentally scattered radiation and rapid determination of the dose rate at test points are important issues in the in-situ calibration of fixed X- and γ-ray radiation dosimeters with open X-ray reference radiation fields. In this study, a Monte Carlo calculation model of the environmentally scattered radiation for in-situ calibration was established, the environmental parameters that may affect scattered radiation under two calibration scenarios were systematically analyzed, and datasets were constructed. Three machine learning algorithms, Support Vector Regression (SVR), Adaptive Boosting (AdaBoost) and Gradient Boosting Regression Tree (GBRT), were used to establish a scattering factor prediction model, evaluate the prediction performance of the model in test sets and experiments, and carry out the application of in-situ calibration of typical dosimeters. The GBRT was found to have better comprehensive performance than the SVR and AdaBoost prediction models did, the GBRT was able to predict the scattering factor on the test set without exceeding the Mean Square Error (MSE) of 1.16E−04, the Root Mean Square Error (RMSE) of 1.08E−02 and the Mean Absolute Error (MAE) of 8.53E−03, respectively, and with R² converging to 1. The maximum relative deviation of the scattering factor in the experiments was −6.9%. This study provides an intelligent method for dose determination in the in-situ calibration of fixed dosimeters, which can be extended to more complex calibration scenarios by expanding the database. At the same time, it provides a feasible idea for replacing isotope radiation sources with X-ray sources.