| Today,when industrial digitization and intelligence are becoming more and more popular,pointer instruments are still an important tool for monitoring the operating status of substations because of their single structure,stable performance,and low price.At present,most analog instruments are manually read instruments,which are affected by subjective factors.With the increase of a single inspection time,the accuracy and efficiency of manual instruments reading will be reduced to varying degrees,this poses hidden dangers to the safe operation of substations.Therefore,Using deep learning algorithms to study the automatic recognition technology of pointer instruments readings has very important practical significance.Aiming at the problems of large on-site environmental impact,poor accuracy and real-time performance of the existing pointer instrument reading recognition technology,we study an automatic recognition technology of substation pointer instruments data based on deep learning.The main research contents are as follows:(1)Research an improved target detection network structure based on YOLOv3 network,and propose a target detection algorithm based on deep learning to solve the problem of automatic detection and extraction of pointer instruments data in the complex background environment of substations.(2)In view of the noise interfence,uneven lighting,and tilt of the instrument images collected by the image acquisition device that hinder the recognition of meter readings,a suitable preprocessing method is proposed.Use Gaussian filtering,Gamma correction,Retinex image enhancement and other algorithms to improve image readability.Mainly,due to the shooting angle,the collected instrument image usually has a certain tilt angle,which is used on the basis of the improved YOLOv3 network structure Mobile Net v3 replaces Dark Net 53 feature extraction network to find key points and use them as feature points of perspective transformation and perform image correction to eliminate dial distortion.(3)Aiming at the rich information contained on the dial,a dial information extraction algorithm based on region segmentation is proposed.The algorithm reduces the amount of data,calculations,and algorithm complexity through image area segmentation.The segmented image is analyzed by connected area,dial information contour extraction,combined with the key point information collected during image correction,to complete the simulation of the center line of the pointer.Combination and the extraction of instrument range dimension information.(4)Based on the angle method,this article uses the reading method of the distance method to study a new calculation method for pointer instruments reading.The method does not need to judge the pointer pointing,and solves the problem that the traditional method is only suitable for uniform scale instruments and not suitable for non-uniform scale instruments.Problems with uniformly scaled instruments.(5)A visualization software is designed to include three modules,such as real-time detection of meter targets,reading of meter indications,and historical data recording and display,through a visual user management interface,To provide managers with intuitive and accurate pointer instrument reading data. |
