Research On Electrical Equipment Identification Algorithm Based On Deep Learning

Electrical equipment is an important part of the power system.The state of electrical equipment will affect the stability of the power system,so it is crucial to keep it running safely and stably.However,the traditional manual inspection methods and video surveillance inspection methods have many drawbacks,such as long working cycle,low labor efficiency,surge in workload,and shortage of personnel,etc.With the rapid development of deep learning in the field of image recognition,applying it to the inspection of power systems can realize automatic identification and positioning of electrical equipment,which greatly improves the speed of identification and ensures the quality of inspection.Therefore,this paper proposes an algorithm based on deep learning for the recognition algorithms of electrical Equipment.The main research contents are as follows:Firstly,an image dataset of electrical equipment is established,which consists of five detection targets,respectively transformers,conservators,fans,mechanism boxes,and insulating sleeve.The methods of image acquisition include shooting with a camera on site and downloading from the Internet.However,training a deep learning network requires a large amount of data,so the collected images need to be augmented.In this paper,data enhancement is carried out by means of linear transformation,balanced histogram,affine transformation and background fusion.Then,the object categories and locations in the dataset are labeled in detail by the labeling software.Secondly,the detection performance of the three target detection algorithms,Fast RCNN,Faster RCNN and YOLOv2,on the above datasets is compared through experiments.Experiments show that the average accuracy of the electrical equipment recognition algorithm based on a Faster RCNN detection model is 78.98%,which is higher than the other two detection algorithms.Therefore,the model is determined as the detection model of electrical equipment due to its relatively good recognition effect.To achieve the goals of high detection accuracy and faster detection speed,the feature extraction network and region proposal network of the Faster RCNN model are improved.The feature extraction network in the original model is VGG16,which is replaced by a deep network ResNet50 with residual structure,to obtain deeper and richer semantic information.After the improvement,the detection accuracy of the model is increased by 3.56%,and the detection speed is two frames per second slower.Then,the improvement method of the region proposal network is to use K-means clustering algorithm.The method reduces the number of anchor boxes,and the size of the regenerated anchor boxes is more suitable for electrical equipment.The detection accuracy of the model is improved,and the convergence speed is accelerated.The improved model detection accuracy is increased by 7.5%compared with the above model,and the detection speed is 5 frames per second faster.The detection accuracy of the overall improved model reaches 90.04%,and the detection speed reaches 11 frames per second.Compared with the original detection model,the detection performance of the improved model is significantly improved,and it can better meet the real-time and accuracy requirements of the power system.