English  |  设为首页  |   加入收藏
 
 
水位图像识别的场景问题处理方法和技术------江海洋,刘林海,李红石
      来源:南京水利水文自动化研究所    [大][中][小]

江海洋 1,刘林海 1,李红石 2

1.浙江省水文管理中心, 浙江 杭州 310009

      2. 杭州海康威视数字技术股份有限公司, 浙江 杭州 310051

摘 要:水位图像识别系统基于人工智能图像识别技术,通过对水尺进行智能识别,直接获得水位的数值,但在实际应用中,由于受场景因素的干扰,稳定性大受影响。通过采用深度学习算法、多帧识别、曝光参数优化,以及摄像机硬件及光学的定制等多种手段相结合,解决各种场景因素的干扰,并在杭州之江水文站进行比测。比测结果表明,智能图像识别水尺系统满足水位观测标准中自记式水位计的要求,和传统水位测量相比具有建设成本低,测量方式高效的优势,具有广阔的应用前景。

关键词:人工智能 ;图像识别;场景处理; 深度学习; 水位; 监测

Scene processing methods and technology for image recognition of water level

JIANG Haiyang1, LIU Linhai1, LI Hongshi2

(1. Zhejiang Province Hydrological Management Center, Hangzhou 310009, China;

2. Hangzhou Hikvision Digital Technology Co., Ltd., Hangzhou 310051, China)

Abstract: The image recognition system of water level is based on the artificial intelligence image recognition technology. Through the intelligent recognition of the stage gauge, the water level value is directly obtained. However, in practic, due to the interference of scene factors, the stability is greatly affected. In this paper, the interference in various scenes is solved by depth learning algorithm, multi frame recognition, exposure parameter optimization, and customizing the camera hardware and optics, etc. The comparative measurement are carried out at Zhijiang hydrometric station in Hangzhou. And the results of comparative measurement show that: the intelligent image recognition system of water level meets the requirements of water level recorders of the standard for stage observation. Comparing with the traditional water level measurement, it has the advantages of low construction cost and high efficiency, and has a broad application prospect.

Key words:  artificial intelligence; image recognition; scene processing; deep learning; water level; monitoring

水位图像识别的场景问题处理方法和技术.pdf (1.49 MB)
2020-03-02 17:32

lixinling2020-03-02浏览(654   [关闭]
   

    

English  |    设为首页  |   加入收藏  

 

 

 
 
进入编辑状态