A research group comprising the Japan Atomic Energy Agency (JAEA) and Nagoya University Graduate School has developed a new method for analyzing radiation measurement data based on machine learning, a field of AI (artificial intelligence).
When using radiation measurements from remote locations to produce distributions for radiation in the environment (radiation mapping), it is necessary to take into account distance from objects that change according to the geography, and radiation attenuation arising as a result of buildings and other objects. Consequently, the dose rate and radioactivity of the locations concerned had been calculated using parameters tailored to the measurement environments.
Following the accident at Tokyo Electric Power Co.’s Fukushima Daiichi Nuclear Power Plant in March 2011, the need for investigating the distribution of radiation in the environment has increased. Radiation detectors are fitted to vehicles, manned helicopters and UAVs (unmanned aerial vehicles) such as drones, and large quantities of radiation measurement data are being obtained in the environment and paired up with location information.
The JAEA attempted to take Big Data constructed using UAV-based radiation measurement data and GPS location information data obtained since the accident at Fukushima, and apply the data to machine learning (deep learning). In doing this, the JAEA used existing software in the machine learning algorithm portion, to make it accessible to anyone to use.
As a result, the JAEA was able to confirm that compared to the conventional method, the new machine learning-based analysis method offers a more than 30% improvement in accuracy (the measurement error when compared against figures for radiation measured on the ground). In addition, it took around an hour to analyze 3,600 items of radiation measurement data using the conventional method, but with the newly-developed method it was found that as a result of having the data learned in advance, the analysis is completed in several minutes.
The research group says that going forward it will aim to further improve the conversion accuracy by adding photograph-based information for identifying structures, topographical information, differences in weather conditions that have a shielding influence on radiation, and so on.
This method have practical applications not only for measuring from the air but also for measuring radiation on the ground and within buildings. It also promises to be applicable to various fields related to the measurement of radiation.