Design And Algorithm Research On Substation Instrument Recognition System

With the development of bionics,machine vision as an important expression form has received extensive attention.However,traditional machine vision mainly based on statistical learning is far less than the requirements of human vision.Therefore,the model with visual as the starting point having both the speed of calculation and the mechanism of visual meets the requirement of biological vision system better.To build a more intelligent substation,more and more inspection robots began to be used for substation equipment testing.In order to improve the ability of the inspection robot,achieve unattended substations,construct intelligent substation,this thesis presents a substation digital display instrument recognition system based on visual models.According to the work requirements of the substation,the function and goal of the substation are determined,and the system design is proposed.Hardware and software components constitute the main content of the system.The hardware part consists of PTZ and camera,the software part consists of two parts: target positioning algorithm design and digital display instrument identification algorithm design.Through the re-selection of the PTZ and the camera greatly improved compactness and flexibility of the hardware platform,and through the improvements of the software part algorithm,the system has a faster computing speed and higher accuracy.Target instrument positioning algorithm is proposed based on the predecessors.After search of the target instrument and calibration of the PTZ error are completed,the latter positioning process is divided into instrument initial positioning and instrument precise positioning.Through judging the position of the instrument to control the PTZ rotation,instrument initial positioning can be realized.Then instrument precise positioning is achieved by camera zooming and focusing.Among this,camera zooming is achieved by judging the size of the instrument to calculate the zooming time.Camera focusing adopted focus depth method based on image processing.Camera auto focus is achieved by combining with visual attention model and color contrast.Digital instrumentation recognition algorithm is the most important part of this study.When looking for digital display instrumentation dial area,the attention model is combined with contour extraction and polygon fitting,greatly improve the positioning accuracy.The canny operator combined with Hough transform and affine transformation can be effective to instrument posture correction and restoration.Pulse-coupled neural network method is used to obtain the binary image.Horizontal vertical projection method can quickly separate and locate the number.When the digital characters are identified,the maximum peak value of the time series corresponding to the pulse-coupled neural network is kept constant after multiple iterations.Through limiting the difference between the maximum peaks of time series of the number to be identified with the number sample less than the threshold,then the instrument digits can be identified.After verification of the above method by experiment,digital display instrument recognition algorithm proposed in this thesis has the characteristics of accurate identification,high speed and adaptability,and the object instrument location algorithm can accurately determine the location of the instrument.