摘要
为了提高国家水资源管控能力灌区计量监控点运行质量,提出一种基于深度学习算法的实时水位识别方法,该方法主要由YOLO-v3对象检测和ResNet刻度识别2种模型构成。通过算法设计、训练和集成,实现与浙江省水资源监控信息平台的集成应用。算法测试结果显示,该方法的测试识别、实际运行准确率和识别响应速度基本达到实际使用需求。随着模型训练数量增加,该方法水位识别进度将继续提高,并具备向检测断面现场环境识别拓展的应用空间。
In order to improve the operation quality of national water resources management and control capacity for monitoring points in irrigation areas,a real-time water level identification method based on deep learning algorithm is proposed in this paper.The method is mainly composed of YOLO-v3 object detection model and ResNet scale recognition model.Through algorithm design,training and integration,the article realizes the integrated application with Zhejiang water resources monitoring information platform.The results of algorithm test and trial run show that the accuracy rate of test recognition,the accuracy rate of actual operation and the speed of recognition response of the method basically meet the needs of actual use.With the increase of the number of the model training dataset,the progress of water level identification will continue to improve,and it has the application space to expand to the field environment identification of detection section.
作者
王磊
陈明恩
孟凯凯
温进化
周鹏程
WANG Lei;CHEN Ming′en;MENG Kaikai;WEN Jinhua;ZHOU Pengcheng(Zhejiang Water Resources Management Cente,Hangzhou 310012,China;Hangzhou Dingchuan Information Technology Co.,Ltd.,Hangzhou 310020,China;Hangzhou Langche Technology Co.,Ltd.,Hangzhou 311100,China;Zhejiang Institute of Hydraulics and Estuary,Hangzhou 310020,China)
出处
《水利信息化》
2020年第3期39-43,56,共6页
Water Resources Informatization
关键词
水尺
水位
人工智能
深度学习
对象检测
图像识别
water gauge
water level
artificial intelligence
deep learning
object detection
image recognition
