一种结合卷积神经网络的水深自动测量方法  

Water Level Estimation Combined with Convolutional Neural Network

作　　者：徐志康 冯径 张之正 舒晓村 XU Zhi-kang;FENG Jing;ZHANG Zhi-zheng;SHU Xiao-cun(College of Meteorology and Oceanography,National University of Defense Technology,Nanjing 211101 ,China)

机构地区：[1]国防科技大学气象海洋学院,南京211101

出　　处：《小型微型计算机系统》2019年第4期793-797,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61371119)资助

摘　　要：针对复杂背景下的水尺定位及水位识别问题,提出了一种结合卷积神经网络的水深自动测量方法.该方法首先采用图像处理技术处理采集的水尺图像实现水尺定位及分割,然后基于模糊C聚类方法分割水尺字符并输入到训练好的卷积神经网络中进行识别,最后构建动态映射算法拟合像素高度与实际高度的映射关系,同时根据测量规则结合卷积神经网络的识别结果确定水深.该算法能够在消除水面倒影的同时提高了测量精度,为补充测量数据和校验水位传感器提供了新途径,在真实环境下采集图像进行识别,最后测量结果能够达到毫米级别,相对误差为0. 5%.To solve the problems of water gauge positioning and water level estimation under complex background,a novel method for automatic measurement of water depth based on Convolutional Neural Network is proposed in this paper. Firstly,we use the image processing technology to process the water gauge image to realize the water gauge positioning and segmentation. Then the fuzzy c-means algorithm is used to segment the water gauge characters and input them into the trained Convolutional Neural Network for identification. Finally,the dynamic mapping algorithm is constructed which mapping the relationship between the pixel height and the actual height. And the water depth is determined according to the measurement rules combined with the recognition results of the Convolutional Neural Network. The water level estimation can improve the measurement accuracy while eliminating the surface reflection,and it provides a new way to complement the measurement data and calibrate water level sensor. The measurement result in a real world image shows accurate detection with the algorithm in this paper,and the final measurement results can reach the millimeter level together with the relative error is 0. 5%.

关 键 词：水位测量 水尺定位 卷积神经网络 字符识别 动态映射 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]