堤防
多光谱图像
遥感
卷积神经网络
环境科学
计算机科学
无人机
杠杆(统计)
地质学
变更检测
人工智能
航测
可扩展性
作者
Mingliang Tang,Dimitrios Zekkos,Dimitrios Zekkos,Kenichi Soga
出处
期刊:
[American Society of Civil Engineers]
日期:2026-03-05
卷期号:: 498-508
标识
DOI:10.1061/9780784486733.047
摘要
The detection of impacted surface soil by animal-burrowing activities is a critical consideration for the performance and longevity of levees. Animal burrows are often detected as part of levee monitoring programs using primarily terrestrial approaches involving visual detection. Geophysical methods have also been successfully used but are still not as common. These techniques require that a person physically walk on the crest, sloping portion, and toe of the levee to detect burrows. Due to the topography, vegetation, and long length of levee systems, detection of animal burrows is not conducted frequently, may not be thorough, is biased towards more easily inspectable areas, and is often incomplete. In this study, a proposed methodology that is based on Uncrewed Aerial Vehicles (UAVs) and multispectral sensors was developed with the goal to detect burrow-impacted soil locations. The methodology was deployed at levees near Yuba City, California, USA. The methodology involves training a convolutional neural network (CNN) model against manually digitized survey results that are based on on-the-ground deployment. The results of the CNN model are compared to the manual labels that delineate the boundaries of burrow-impacted soil spots to assess the algorithm’s prediction accuracy. It is shown that the UAV-enabled methodology combined with the predictive model can leverage information collected from ground surveys to improve monitoring effectiveness. The data analysis is shown to be easily scalable and provides an efficient way to detect burrow-impacted soil over large areas.
科研通智能强力驱动
Strongly Powered by AbleSci AI