Research And Implementation Of Power Equipment Identification Based On Convolutional Neural Networks

The normal operation of power equipment is the foundation of long-term safety and stability of power system,therefore,regular inspection of power equipment is very important.In the past,the inspection methods have various shortcomings,which result in the waste of manpower and material resources,and the effect is not very ideal.In recent years,with the gradual maturity of artificial intelligence and computer vision technologies,its application to the detection of the running state of power equipment can make up for the shortcomings of the traditional methods.This method has a long-term development prospect.In this paper,a new system for automatic detection and recognition of power equipment operation state is proposed based on the knowledge of convolutional neural network and computer vision.The system can automatically detect the position of power equipment and distinguish its running state,and automatically read the pointer instrument.The main research contents and work of this paper are as follows:Firstly,an algorithm for automatic detection and recognition of power equipment based on convolutional neural network is designed and implemented.The Faster RCNN+ResNet101 model was used to complete the recognition of power equipment on a high-performance computer.At the same time,MobileNet+SSD model was designed and implemented for the raspberry PI embedded device with weak computing power and portability.In addition,we made a data set for the detection and recognition of power equipment,and respectively used two models to conduct training and testing on this data set.The Faster RCNN+ResNet101 model has achieved high recognition accuracy,and it can barely reach the real-time requirement in the detection speed.The MobileNet+SSD model with neural network computing stick on raspberry pie slightly decreased in recognition accuracy,but the detection speed reached 15fps,fully meeting the real-time requirements of the systemSecondly,an algorithm of automatic reading is designed and implemented for the pointer instrument commonly used in power system.With the knowledge of image processing,the image of pointer instrument is firstly processed with grayscale,histogram equalization and image denoising.Then the image was morphologically inflated and Canny operator was used to detect the edge information of the image.In addition,Hough transform is used to detect the position of the line where the pointer is located in the edge information.At last,the Angle method is used to automatically read the line where the pointer is located.Thirdly,an automatic detection,identification and monitoring system of power equipment based on convolutional neural network is designed and implemented from the aspects of software and hardware.Different hardware devices are matched with corresponding software algorithms to meet the use requirements of the system in different situations.At the same time,the front-end interface is designed for the system,and various functions of the system are integrated to provide a more humanized human-computer interaction process.