| Today,electric energy is one of the most important energy sources in people’s production and life.The stability and reliability of its power system determine whether the entire power grid can operate safely and efficiently.Fault detection of transmission lines is the basis for ensuring the stable operation of the power system,and it is also an important part of power inspection and monitoring.With the rapid development of UAV technology,UAV inspections of power systems have gradually replaced traditional manual inspections.With the rapid development of UAV technology,UAV inspections of power systems have gradually replaced traditional manual inspections.On the other hand,with the development of big data and artificial intelligence technology,feature recognition technology has been rapidly improved.Intelligent fault detection of power grid transmission lines using UAV inspection images has become an important part of the intelligent and information management of the power system.Based on the technical background of deep learning,this article aims at the problems of small number of fault samples,poor image quality,and low target detection accuracy in practical applications,carries out the construction of the UAV inspection image data set of the power system,the research of the lowilluminance image enhancement method based on deep learning,and the research of the power system fault intelligent detection technology based on the improved Eifficient Det-D0,and build the evaluation index system,and carry out the experimental verification.The main research work is as follows:(1)Construction of UAV inspection image data set for power system.In response to the need for a large number of high-quality annotated image data during the actual training of the fault detection network,on the one hand,the public data set is downloaded from the network to obtain aerial images containing normal and defective insulators;on the other hand,drones are used to collect at different time points Aerial image of high-voltage transmission lines in Heilongjiang.In addition,an image generation network based on Style GAN2 and a discriminant network based on residual structure are selected,using the idea of generating confrontation,and using existing image data as training input to further generate the target image.Use the image annotation tool to mark the fault of the acquired image.Then,through operations such as image zooming,flipping,distortion,and increasing image noise,the image is enhanced in the data layer to increase the diversity of image features,reduce the model’s requirements for data completeness,and provide the subsequent transmission line fault detection network data support.(2)Research on Low Illumination Image Enhancement Method Based on Deep Learning.Aiming at the problem of difficult target feature extraction and low detection accuracy in low-illuminance images obtained by drone inspection,at the same time,in order to achieve end-to-end image enhancement,low-illumination image enhancement methods are studied.First,the image illuminance classification network is designed based on the improved VGG network,and the network structure is lightened by cutting the network width and depth and introducing perforated convolution.Set the probability threshold,and input the low-illuminance image into the network to realize the illumination classification of the image.Then,study the structure design method of image enhancement network based on strong supervision and weak supervision training method,and complete the training and test of image enhancement network based on Retinex-Net and DCE-Net by designing different loss functions.Finally,an image enhancement effect evaluation system is constructed to realize the quantitative evaluation of low-illuminance image enhancement performance.(3)Research on power system fault intelligent detection technology based on improved Eifficient Det-D0.Based on the UAV inspection image data set of the power system,research on the intelligent detection technology of transmission line faults is carried out.In order to further improve the target detection performance,based on the single-stage target detection network Eifficient Det-D0,the original network framework is improved,and the parameters of each layer of the prediction branch network are given.Choose Focal Loss and Smooth L1 as the loss function of the network,and set the hyperparameters based on the experience of classic network parameter setting.Considering factors such as the stability of gradient update and the efficiency of backpropagation,the Adam optimizer is selected for parameter optimization.In order to better match the target of the data set,the prior frame height and width values are updated based on the K-value clustering method,and the learning rate adjustment mechanism based on cosine annealing and the training method based on Mosaic data enhancement are used to train the fault detection network.Finally,intelligent detection of transmission line faults is realized.(4)Experimental verification of intelligent detection algorithm for transmission line faults.Based on the improved Eifficient Det-D0 intelligent transmission line fault detection network and patrol inspection test collection images,the experimental verification of the transmission line fault intelligent detection algorithm is carried out.Use the confusion matrix to construct a fault detection accuracy evaluation system that includes performance indicators such as detection accuracy,recall,F1 index,m AP and carry out verification experiments for different detection target categories,different image illuminances,and whether to improve the network structure,and statistics Calculate the evaluation index under different experimental conditions.The experimental results show that improving the network structure,enhancing the image illuminance,and optimizing the training method can effectively improve the fault detection accuracy,and can provide scientific basis and technical support for the intelligent fault detection and recognition in the actual transmission line inspection process. |
PDF Full Text Request