Canal discharge identification is essential for irrigation water metering, water management, and ecological protection. With the rapid advancement of UAV aerial photogrammetry, UAV-based large-scale field flow field observations have emerged as a prominent research focus. Given the capability to accurately retrieve canal surface flow velocities, a critical challenge remains in UAV remote sensing-based canal hydrological monitoring: how to identify key cross-sections, obtain high-resolution surface flow field information, and enable timely canal cross-section discharge estimation during sudden flood events. To address the aforementioned challenges, this manuscript combines deep learning algorithms with Kalman filtering and monocular ranging techniques. Artificial square sheet tracers are released into the canal, based on which a YOLO-DeepSort deep tracking framework is constructed. Based on the established UAV-based canal flow velocity perception platform using deep learning, this manuscript achieves mAP@0.5 of 0.995, with precision and recall both reaching 1.0 for real-time tracer detection and flow velocity identification via UAV low-altitude remote sensing. The average relative error of velocity estimation is within 7%, and discharge inversion errors are 1.7%, 6.4%, and 4.6% for the three canal sections, respectively. The surface flow field and cross-sectional velocity distribution of the observed sections are obtained accurately. This manuscript is expected to provide a systematic scientific basis for UAV low-altitude remote sensing-based canal discharge monitoring.
Z et al. (Mon,) studied this question.