Creating radiation maps smarter with AI
—Harnessing big data on radiation measurements from Fukushima—
Until now, analyzing radiation measurement data and estimating the distribution of radiation sources have required analysis based on a wide range of information, all of which took a lot of computing time.
In the wake of the nuclear disaster at Tokyo Electric Power Company’s Fukushima Daiichi Nuclear Power Plant (hereafter, “1F”), work conducted to measure the distribution of radiation throughout the environment (radiation maps) has resulted in vast quantities of data (hereafter, “big data”) stored as radiation measurement results.
The Japan Atomic Energy Agency has worked with Nagoya University to successfully develop a new method for analyzing radiation measurement data, by harnessing the capabilities of machine learning—a type of AI (artificial intelligence)—to create radiation maps quickly and accurately using big data acquired from unmanned aerial vehicles (“UAVs”).
Using this method allowed radiation maps of radiation measurement values on the ground to be created with 30% or better accuracy compared to previously used methods. Analysis work that used to take one hour or longer in the past was completed within some minutes.
This method will allow detailed radiation maps of the 1F evacuation zone to be created quickly and accurately, and is expected to be useful as a scientific basis for decontamination work or lifting evacuation orders.
Future efforts will be made to further increase accuracy through means such as adding information for identifying buildings based on photos or differences in weather conditions.
This method is not limited to UAVs, but is also applicable for radiation measurements in a broad range of environments. Other applications are also anticipated, including radiation measurements for medical purposes or for calibrating detectors.