2022 年 71 巻 7 号 p. 631-636
In agricultural canals, the plate thickness of steel sheet piles decreases locally in the tidal zone and the cross section is lost. Degradation of structural performance can lead to the occurrence of sudden accidents. It is necessary to evaluate the corrosion condition and thickness of steel sheet piles by periodic investigation. In this paper, the thickness of the steel sheet piles is evaluated by machine learning of digital images (RGB images and infrared images) taken by UAV. As a result, the random forest algorithm can be used to determine whether the plate thickness of the steel sheet piles is greater than the corrosion allowance or not (accuracy: 0.828). A particularly useful parameter is the difference value of infrared temperature between 14:00 and 17:30. Therefore, the plate thickness of steel sheet piles can be evaluated by acquiring infrared images at two time points with large temperature differences.