Research And Application Of Reading Algorithm For Complex Water Level Scales

With the continuous development of society and the improvement of water conservancy facilities,the importance of water level gauge readings in tasks such as flow regulation,water level control,hydroelectric power generation,flood control and irrigation is increasing.The water gauge and gate gauge laid along the canal in the South-North Water Transfer Project are two different types of water gauge,and the reading results have important practical value for local administrations and bureaus.In engineering practice,the camera takes two kinds of water level scale scene images,the computing equipment inputs image data into the algorithm,and the algorithm outputs the final reading results.This thesis focuses on algorithm research and system application of image reading tasks in the water gauge and gate gauge scenarios of the South to North Water Diversion Project,with a focus on research and experimentation of reading algorithms.At present,the water level reading technology in the above two scenarios is relatively backward,relying on human eye readings and traditional image processing algorithms cannot meet the needs of accuracy and real-time in engineering.In addition,the occurrence of abnormal conditions such as foggy weather and night places higher requirements on the robustness of the algorithm.This thesis conducts algorithm research and system application for two scenarios of water gauge and gate gauge,mainly including:(1)A water gauge level detection algorithm based on keyword processing and multiple water surface positioning has been proposed.The algorithm flow described in this thesis involves image preprocessing,edge detection,morphological processing,Affine transformation,depth learning,core reading analysis and other steps.At present,the water gauge water level detection algorithm has been deployed in the intelligent water level detection system of the South-North Water Transfer Project.The landing accuracy error is not more than 2cm,and the speed can reach 22.2fps,which has high industrial application and theoretical research value.(2)A water level detection algorithm for gate rulers based on gate ruler detection and ruler post-processing has been proposed.To improve the accuracy of the detection process,this thesis focuses on solving the problem of uneven number of cursor and scale samples in the training samples.Due to the complexity of the application environment,this thesis applies a two-stage keyword detection algorithm.In addition,to analyze the final reading based on the detection results,this thesis proposes a highly robust core reading analysis strategy.At present,the water level detection algorithm for the gate ruler has been successfully deployed in the system,with a landing accuracy error of no more than 1.6dm and a speed of up to 17.2fps.Compared to other algorithms,this algorithm is the first to conduct research on the reading algorithm of a gate ruler with a double ruler surface,which has significant reference significance for reading tasks in similar scenarios.(3)Introduced the South to North Water Diversion Project intelligent water level detection system.The detection system designed and developed based on the system hardware and software framework is more adaptable to algorithms and exhibits robustness in long-term operation.Through algorithm research and system application,the task of water gauge and gate gauge reading of the South-North Water Transfer Project was successfully completed,and the final algorithm delivered well met the requirements of accurate real-time reading of complex water gauge along the canal in the South-North Water Transfer Project.In addition to performing well in engineering applications,the algorithm proposed in this thesis utilizes a combination of traditional and deep learning methods to perform high-precision real-time processing of video streams,and performs special processing on abnormal situations such as severe weather,enhancing robustness.In addition,the scale post-processing algorithm designed and applied in the brake scale scenario is the first of its kind in the dual scale vernier scenario,which has significant reference value for reading tasks with dual scale surfaces.In summary,the complete complex water level scale reading algorithm and intelligent water level video detection system proposed in this thesis have both theoretical and practical value in the fields of image reading and engineering practice.