基于深度学习的水管式沉降仪数据采集方法

A Data Acquisition Method for Water Pipe Settler Based on Deep Learning

传统水管式沉降仪主要通过压力传感器来测量水位进而计算坝体内部沉降量,其存在测量稳定性差的问题。针对此弊端,提出一种基于深度学习的水管式沉降仪数据采集方法,包含特制水尺及相应算法。通过摄像机从上至下分段拍摄特制水尺并筛选出水位所在图像;采用二维码标识对图像进行角度校正并切割出水尺及测量管区域,同时读取厘米级水位坐标;基于UNet搭建图像抠图模型精确分割测量管中水位;最后结合水位线坐标,采用YOLOv5模型识别标尺毫米级坐标。试验结果表明,该技术环境适应性强,测量结果与人工读值相比,其误差均小于0.3 mm,满足实际工程的需要。

Traditional water pipe settlement meters mainly measure water levels through pressure sensors to calculate the settlement within the dam.However,it has the problem of poor measurement stability.To address these shortcomings,a water level measurement method for water pipe settlement meters based on deep learning is proposed.This technology includes a specially designed water gauge and corresponding algorithms.The method involves segmenting the water gauge from top to bottom using a camera and selecting the image containing the water level line.A QR code is used to correct the angle of the image and separate the water gauge,simultaneously,and centimeter-level water level coordinates are obtained.Based on the UNet model,an image matting model is used to accurately segment the water level.Finally,a YOLOv5 model is employed to identify millimeter-level coordinates on the gauge.Experimental results demonstrate that this technology has strong environmental adaptability.The measurement error is less than 0.3 mm compared to manual readings,meeting the requirements of practical engineering applications.