Identification And Location Of Substation Safety Monitoring Image Based On Convolutional Neural Network

With the development of computer ability and artificial intelligence,the society has put forward higher requirements for the intelligence and automation level of power system.It is hoped that real-time visualization and intelligent judgment of personnel and equipment risks can be realized in various types of operation scenarios such as power transmission,substation,power distribution,infrastructure construction and so on.For example,in the daily monitoring work of the substation,a large number of infrared images of the substation equipment and the safety helmet images of the staff have been accumulated.The staff need to determine the category of the substation equipment,determine its operation status,judge its fault form,and judge whether the behavior of many people in many places complies with the safety rules and regulations.The influence of various subjective and objective factors,such as the image quality,the variety of substation equipment and personnel behavior,has brought great difficulties to the safety supervision of substation,resulting in the low efficiency and low degree of automation of substation safety supervision.Aiming at the monitoring problem of substation safety monitoring image,this dissertation applies deep learning and image processing technology to intelligent recognition of substation safety monitoring image,deeply studies image preprocessing,object recognition and location methods such as image enhancement,image denoising and feature selection,and puts forward the object recognition and location method of substation safety monitoring image based on Convolutional Neural Network.According to the characteristics of infrared images of various transformer equipment in the substation,this dissertation proposes a preprocessing method of infrared image of transformer equipment,which combines Wavelet Transform,MSRCP(Multi Scale Retinex with Chromaticity Preservation)image enhancement method,Bilateral Filter,Threshold Segmentation and Feature Selection,so as to realize the segmentation and extraction of substation equipment in infrared image and reduce the information of non-target area of infrared image.Then,GIOU(Generalized Intersection-Over-Union)is introduced to improve the Convolutional Neural Network,YOLOv3,such as non maximum suppression algorithm and anchor optimization.At the same time,image enhancement technology and transfer learning strategy are used to train the improved YOLOv3 network for small dataset get the final model.The experimental results show that the proposed method based on improved YOLOv3 has the characteristics of high accuracy and high detection speed,which provides a good basis for further fault diagnosis.According to the characteristics of the safety helmet image of substation personnel,this dissertation uses MSRCP image enhancement method and Bilateral Filtering to preprocess the safety helmet image of substation personnel.Then,the K-means++ algorithm based on GIOU is used to improve Faster R-CNN.The experimental results show that the method based on the improved Faster R-CNN makes up for the defects of the improved YOLOv3 algorithm in small object or multi-target detection,and effectively improves the accuracy of the identification of substation personnel safety behavior.The method can meet the real-time monitoring requirements of electric substation,and can improve the efficiency and accuracy of online monitoring and improve the level of automation intelligence of substation.