Research On River Velocity Measurement Based On Video And Image Recognition

The flood disaster has a great threat to the safety of the people's life and property,and the flow and flow measurement of the river during the flood season have an important effect on the flood warning.The rural hydropower stations in mountainous areas are often inconvenient and relatively backward in technology.The traditional artificial velocity measurement method is lack of convenience during the flood season,and at the same time,the safety of the surveyors can not be guaranteed.Therefore,the development and design of a flow velocity measurement method suitable for field unattended environment is of great significance for flood control and early warning in remote areas.In order to break the limitations of traditional flow measurement methods and meet the requirements of timeliness,convenience and safety of flood warning,a method based on video and image recognition was proposed to measure the flow velocity.The video of the target river surface is taken by the network HD smart camera,and the trained convolution neural network model is used to identify the watermarks of the video image,thus the flow velocity information is output.This method is a non-contact flow measurement method,without manual delivery of buoys or other tracers,which can meet the requirements of river flow monitoring under unattended environment in remote areas.In addition,in order to train a suitable convolution neural network model,a large number of river video samples are collected,and a method called Data Representation Clustering is proposed,which reduces the difficulty of data set annotation.The main work of this article is as follows:(1)A river flow video image based pattern recognition method was proposed to measure the flow velocity of the water.According to the differences in the texture characteristics of the river surface and the differences in the characteristics of the fluid before and after the difference in the flow velocity,video images are captured by the camera,and the velocity is classified after the characteristics are identified,thereby achieving the purpose of dividing the velocity class.(2)The Data Representation Clustering which incorporated low rank representation,similarity learning and cluster structure constraints into the same framework is proposed and applied to the clustering marking of water flow image data sets.It reduces the huge workload for measuring the flow rate of a large number of water flow image data sets and calibrating them.(3)The convolution neural network is used to classify the flow image and video frame difference classification,and the difficulty of the model training is reduced by migration learning.The two-stream convolutional neural network is constructed to classify the flow velocity range of the flow video frame difference and the image.The experimental results on the flow data set show that the convolution neural network model has a higher recognition rate than the traditional machine learning classification algorithm,while the two-stream convolution neural network model further improves the accuracy of the algorithm based on a single video stream or a single image stream convolutional neural network model.(4)A river flow velocity measurement system has been designed and developed.The system has been implemented in the Jiu Gong Shan cascade Small Hydropower Group,and the effectiveness and real-time performance of the system has been tested in the field.The final results show that the flow measurement method based on video image is convenient,high recognition rate and good real-time performance,and can meet the basic requirements of water flow velocity measurement and flood control of hydropower stations.