Design And Implementation Of Intelligent Grid Inspection System Based On Image Recognition

With the continuous improvement of the national industrial and economic strength,the demand for electricity has increased rapidly,which in turn has promoted the rapid development of the power system.The area covered by the power system has been continuously expanded,and the grid structure has become more and more complex.This has also brought a severe challenge to workers’ inspection work.Substation op-erators have always used the standard paper card for operation guidance,which is not only inconvenient to carry,but also has low efficiency.It requires a lot of time to be consulted on site.How to increase the efficiency of power work by introducing new technologies and strengthen the safety factor of power work have become a hot topic in the research and practice of related fields.In this paper,we focuses on how to imple-ment the intelligent grid inspection system from three aspects including overall design,image recognition method design and the specific implementation of the system.The main research contents of the paper include:Firstly,this article starts from the actual grid inspection task and details the in-door and outdoor inspection work of the staff.Then,we abstract the traditional purely human-reliant task into a computer-understandable work-flow,and converted the core patrol contents of the task into the problem of computer vision:the switch status recog-nition problem of the power protection screen and the problem of meter reading.Secondly,for the power screen equipment in substations,this paper introduces a method for detecting and recognizing the status of the switch based on image semantic segmentation and K-means clustering.It classifies pixels by using a fully connected neural network and a dense conditional random field model,and then use morpholog-ical and other image processing methods to further reduce the noise in the semantic segmentation results.After that,we use the pixel accumulation histogram to cut the screen image row by line to obtain a single switch image.Then we use the K-means clustering algorithm to find the cluster centers of each of the switch,and consider the geometric positional relationship between the switch to determine the status.Next,this paper introduces a method for pointer meter detection and reading esti-mation based on deep learning and image processing in substation scenes.We use fully convolutional network(FCN)to semantically segment the image,and then use affine transformation to correct the inclination of the image caused by nonstandard shooting,and finally obtain the location of the pointer meter in an image.Then,the histogram equalization and filtering of the obtained instrument image are performed to reduce shadow interference and enhance the image area of the meter to be identified.Then,according to the idea of Hough transform,we propose an improved Hough transform method to obtain the angle of the pointer.And we calculate the reading of the pointer meter based on the angle.Finally,through the system construction scheme introduced in this paper,we im-plement the identification method described in the paper,and use the images we have collected in the actual substation environment to verify the correctness and robustness of the method.