2021年福島県沖の地震による被害調査への深層学習活用型画像診断システムの適用および実用化のための基礎的検討

抄録

In this study, we applied a diagnostic imaging system, which is an upgraded version of a previously proposed diagnostic imaging technology, to the actual damage inspection of timber houses damaged in the 2021 Fukushima earthquake. We also examined the relationship between pixel resolution and damage detection accuracy as a basic study of the imaging method, with an eye to self-diagnosis by occupants using smartphones. The results of damage detection and damage rate calculation using images with a pixel resolution of 0.3 mm/px or less showed that the error in the image diagnosis results was 0.133%, compared to the maximum deviation of 0.567% (assumed to be a permissible error) in the visual measurement by three persons, indicating that the damage rate could be calculated with the same or higher accuracy as the visual measurement.