Research On Automatic Reading Of Substation Pointer Instrument Based On OpenVINO And Deep Learning

Electricity instrumentation is an essential tool for monitoring and recording the safe and efficient operation of the power system,as electrical instrumentation is often unable to transmit data,according to the traditional manual method of reading and recording pointer instrumentation data,making a serious waste of human resources,and the monotonous repetition of work content to make the inspection staff low efficiency,easy because of inattention or visual fatigue and other reasons for the wrong inspection or omission,in addition to the harsh environment and even endanger the personal safety of the inspection staff.In recent years,with the development of substations towards intelligence,less manned mode and unmanned mode,more inspection robots have been put into the production of substations,and one of their main functions is to collect images of various electrical instruments in substations.However,how to efficiently and accurately read the instrument image data acquired by the inspection robot has become an important research issue at this stage.To address this,this thesis conducts the following research based on Open VINO and deep learning algorithms and combines the characteristics of substation instrumentation images to analyze and solve the problems faced by the automatic instrumentation reading process.(1)Faster R-CNN-based pointer instrument dial recognition method is proposed to extract instrument dial regions from pointer instrument images taken by inspection robots using Faster R-CNN target detection technique.Firstly,we train the pointer instrument detection model by instrument image dataset,and then optimize and accelerate the model by Open VINO tool to improve the instrument detection accuracy and speed,so that the model can be embedded into Inter multi-platform hardware and realize the accelerated inference calculation of the model,which can be combined with the substation inspection robot to perform efficient inspection work.(2)An improved Mask R-CNN-based image segmentation of substation pointertype instrument dial tick marks and pointers is proposed.Since the image segmentation effect of the original Mask R-CNN network is slightly less accurate when the target is an irregular image,we improve the Mask R-CNN network so that the segmentation image accuracy of the instrument dial scale line and pointer can be improved,thus improving the accuracy of automatic reading of the pointer instrument.(3)A correction method of instrumentation based on linear transformation theory is proposed.If the scale line outline is found to be distorted after image segmentation,the correction of the dial area of the pointer instrument is carried out.The distorted instrument dial is transformed into a circular instrument dial by the perspective transformation method,and then the deviation of the image angle of the instrument dial after the perspective transformation is compensated by the affine transformation method,and the accurate reading of the pointer instrument can be calculated by the angle method.(4)We propose to design an automatic reading system for pointer instruments by substation inspection robots based on Open VINO,deploy the system to a Raspberry Pi board as an edge detection device for inspection robots,and send the instrument dial detection results to the terminal through a network protocol.The system can realize substation pointer instrument image detection,image segmentation,image correction,calculate and identify instrument readings and display the reading results in real time,and finally record the pointer instrument reading data to the inspection log.In order to verify the optimization acceleration effect of the Open VINO tool and the accuracy of the automatic meter reading system studied in this paper,we use a large number of real substation meter images for experimental analysis: firstly,the same model optimized by Open VINO model and without Open VINO model is experimentally verified,and the experimental results show that the model optimized by Open VINO model reduces the inference time by about 40 times.The experimental results show that the inference time of the model optimized by the Open VINO model is shortened by about 40 times;secondly,the experimental results show that the improved Mask R-CNN network can effectively improve the accuracy of meter reading,which can reach 97.58%with an average relative error of 0.35%,meeting the requirements of the inspection robot of substation for the inspection of pointer meters.The requirements.