期刊
JOURNAL OF HYDROLOGY
卷 626, 期 -, 页码 -出版社
ELSEVIER
DOI: 10.1016/j.jhydrol.2023.130233
关键词
Particle image velocimetry; Surface flow velocity; Image analysis; River monitoring; ANN; Soft computing
River flow monitoring has seen rapid development thanks to advancements in non-intrusive optical methods. These methods record and analyze the displacement of floating tracer materials to obtain surface velocity fields. By coupling this data with geometric information, river discharge can be assessed. The accuracy of optical techniques is influenced by various factors, such as environmental conditions and software parameterization. This study proposes an automatic procedure for identifying and extracting the most suitable portion of recorded videos, improving the performance of LS-PIV software. The procedure is implemented using an Artificial Neural Network trained with data collected from different rivers in Sicily, and its effectiveness is evaluated based on error in reproducing surface velocity profiles.
River flow monitoring has recently experienced rapid development due to advancements in optical methods, which are non-intrusive and enhance safety conditions for operators. Surface velocity fields are obtained recording and analyzing displacements of floating tracer materials, artificially introduced or already present on the water surface. River discharge can be assessed coupling the surface velocity fields with geometric data of a cross section. The accuracy of optical techniques is strongly affected by different environmental and hydraulic factors, and software parameterization, with tracer features that often play a prominent role. An adequate density and spatial distribution of tracer is required to ensure a complete characterization of surface velocity fields. In practical applications such conditions might occur only for a limited portion of the entire acquired images sequence. This work proposes an automatic procedure for identifying and extracting the best portion of a recorded video in terms of seeding characteristics and demonstrates how LS-PIV software performances can be enhanced through this approach. The procedure is implemented through a data-driven empirical approach based on an Artificial Neural Network, trained using data collected during an extensive measurement campaign across different rivers in Sicily (Italy). Performances are evaluated in terms of error in reproducing surface velocity profiles along specific transects, where benchmark profiles derived using an Acoustic Doppler Current Profiler are available. The procedure, also tested via numerical simulations on synthetic image sequences, outperformed an approach based on an existing metric for seeding characterization and represents a simple and useful tool for LS-PIV based applications.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据