Determination of radiation source direction is important for rapid response in radiation monitoring and security applications. Because the count distribution measured by multiple detectors varies nonlinearly with source direction and shielding conditions, data-driven approaches can be effective for directional estimation. This study experimentally verified the concept of an artificial neural network (ANN)-based directional radiation monitoring system. The proposed system consists of four 3″×3″ NaI(Tl) detectors arranged in a square configuration with wall-type collimators to enhance direction-dependent count-ratio patterns. Detector count ratios were used as input features for an ANN regression model trained with a simulation-based database generated using Monte Carlo simulation. Experimental measurements were conducted with a 137 Cs source at 6 m over source directions from 0° to 90°. The experimental count-ratio patterns agreed with simulation results, supporting the validity of the simulation-based training data. The ANN model achieved a mean angular error of 5.55° and a maximum error of 11.32° for the experimental dataset, although the 6 m condition was outside the training distance range of 50–400 m. In addition, performance became relatively stable when total counts exceeded approximately 500. These results demonstrate the feasibility of the proposed system and support further development of directional radiation monitoring systems.
Chung et al. (Fri,) studied this question.