基于改进YOLOX的城市河道智能水位测量算法

Intelligent water level measurement algorithm for urban rivers based on improved YOLOX

针对目前基于深度学习水位测量算法存在特征信息提取不充分问题,提出一种基于改进YOLOX的城市河道水位智能测量算法。为了提高YOLOX对多类别密集目标的识别率,在特征融合网络中引入CBAM注意力机制,并采用基于计算目标框信息的损失函数D-IoU加快模型收敛。该算法利用改进后的YOLOX对水尺刻度进行识别与统计,并计算出水位值。试验表明提出的新算法对水尺刻度和数字的平均识别率分别达98.62%和92.23%,最终计算水位的平均误差为1.16 cm,较其他图像识别水位测量算法的平均误差减少了1.76 cm,可实现高精度智能测量城市河道的水位值。

In response to the problem of insufficient feature information extraction in current deep learning based water level measurement algorithms,an intelligent water level measurement algorithm for urban rivers based on improved YOLOX is proposed.To improve the recognition rate of YOLOX for multi-class dense targets,CBAM attention mechanism is introduced in the feature fusion network,and a loss function D-IoU based on calculating target box information is adopted to accelerate the convergence of the model.This algorithm uses the improved YOLOX to identify and statistically analyze the scales and numbers on the water gauge,and calculate the water level value.The experiment shows that the proposed method has an average recognition rate of 98.62% and 92.23% for water level scale and number,respectively.The final average error in calculating water level is 1.16cm,which is 1.76cm less than the average error of other image recognition water level measurement algorithms.It can achieve high-precision intelligent measurement of water level values in urban rivers.