Autonomous UAV piloting using a total station and deep learning for dam body degradation surveys

For dams in narrow mountainous areas, the accuracy of satellite positioning is uncertain. As such, we have applied UAV self-positioning estimation technology with a total station to enable high-precision autonomous flight. By aerially photographing the entire dam body surface uniformly in high resolution from an equal head-on distance, this makes it possible to capture image data useful for AI image recognition. Using deep learning on the huge amount of captured image data, image recognition detects deterioration information automatically. By manually marking degraded sections in some of the captured images, the deterioration detection AI can learn how humans detect deterioration, enabling it to automatically detect deterioration in the entire dam body and saving work in dam body surveys. The detection results can be used in combination with photogrammetry techniques to calculate position and magnitude of deterioration and quantitatively assess deterioration in each sector of the dam body to serve as basic data for dam health assessments. With continuous surveying over multiple years, results can also be compared over time.