基于深度学习的多级水尺水位视觉测量方法OA

Compared with radar,ultrasonic,and other non-contact water level gauges,image recognition-based staff gauge measurement technology has the advantages of oblique detection,intuitive results,and no temperature drift.In recent years,it has been increasingly applied to single-stage vertical staff gauge measurements.However,for short-pile multi-level staff gauges commonly deployed on river banks with wide and gently inclined slopes,there are still challenges such as low inspection efficien-cy,complex system calibration,and limited detection accuracy for distant small targets and water lines.To address these issues,this paper proposes a water level visual measurement method that combines multi-level staff gauges detection with waterline re-gression.A bank-based online measurement system is developed using a high-definition fixed-focus camera.For object detec-tion,a YOLOv5 model with the CBAM attention mechanism and an additional small-object detection head is adopted.Mean-while,an SE-HRNetS waterline regression model with multi-scale feature fusion and the SE channel attention mechanism is de-signed to improve the waterline detection accuracy of low-resolution images.To verify the performance of the proposed method,comparative experiments under different lighting and flow conditions were carried out at Pengfang Hydrological Station in the Gan-jiang River basin.The results show that the system achieves an average precision of 94.82%on the experimental station dataset,and the comprehensive uncertainty of water level measurement under day-night and flood scenarios is less than 1.72 cm.The proposed method provides an intuitive and efficient solution for visual water level measurement in small and medium-sized rivers with wide and gently sloping cross-sections.