Experimental study on intelligent detection for surface defects on earth-rockfill dams based on UAV images and TLS point clouds

This study proposes an intelligent detection solution for surface defects on earth-rockfill dams, including unmanned aerial vehicle (UAV), terrestrial laser scanning (TLS) and Deep Convolutional Generative Adversarial Network (DCGAN)-based defect intelligent detection method. To this end, an earth-rockfill dam simulation platform is constructed to explore the defect detection performance of UAV and TLS on earth-rockfill dam, and the impacts of various factors are evaluated by changing the ground control point arrangement or weather scene. Additionally, a DCGAN-based method for generating defect images is proposed for data augmentation of limited on-site images, which provides a new approach for the pain point of over-reliance on datasets for object detection in practical engineering. By learning from small batches of defect images, plenty of similar images are generated. The results indicate that integrating UAV and TLS enables high-precision and rapid detection of surface defects on earth-rockfill dams. The defect images generation method effectively improves the detection accuracy of object detection model under limited samples.