Research on Intelligent Crack Detection in a Deep-Cut Canal Slope in the Chinese South–North Water Transfer Project

Abstract

The Chinese South–North Water Transfer Project is an important project to improve the freshwater supply environment in the Chinese interior and greatly alleviates the water shortage in the Chinese North China Plain; its sustainable, healthy, and safe operation guarantees ecological protection and economic development. However, due to the special expansive soil and deep excavation structure, the first section of the South–North Water Transfer Project canal faces serious disease risk directly manifested by cracks in the slope of the canal. Currently, relying on manual inspection not only consumes a lot of human resources but also unnecessarily repeats and misses many inspection areas. In this paper, a monitoring method combining depth learning and Uncrewed Aerial Vehicle (UAV) high-definition remote sensing is proposed, which can detect the cracks of the channel slope in time and accurately and can be used for long-term health inspection of the South–North Water Transfer Project. The main contributions are as follows: (1) aiming at the need to identify small cracks in reinforced channels, a ground-imitating UAV that can obtain super-clear resolution remote-sensing images is introduced to identify small cracks on a complex slope background; (2) to identify fine cracks in massive images, a channel crack image dataset is constructed, and deep-learning methods are introduced for the intelligent batch identification of massive image data; (3) to provide the geolocation of crack-extraction results, a fast field positioning method for non-modeled data combined with navigation information is investigated. The experimental results show that the method can achieve a 92.68% recall rate and a 97.58% accuracy rate for detecting cracks in the Chinese South–North Water Transfer Project channel slopes. The maximum positioning accuracy of the method is 0.6 m, and the root mean square error is 0.21 m. It provides a new technical means for geological risk identification and health assessment of the South–North Water Transfer Central Project.

Keywords

• central line project of south-to-north water diversion
• imitation ground flight
• ultra-high resolution
• deep learning
• intelligent detection
• field location