Improved YOLOx-based Transmission And Distribution Line Equipment Detection And Defect Identification

The scale of Chinese power system is huge,the complexity of the system is high,and the need for safety monitoring is outstanding.Due to the complex terrain and wide distribution of transmission and distribution lines,the equipment categories and defects to be inspected are numerous and vary greatly in size,and the traditional inefficient manual inspection can hardly meet the daily line inspection demand,so the State Grid and Southern Power Grid introduce UAVs for intelligent inspection of transmission and distribution lines,and for problems such as difficulties in detecting power equipment and defects in UAV inspection images,this paper combines the advantages of deep learning based on YOLOx,the equipment detection and defect identification method of distribution lines and transmission lines is proposed to provide technical ideas for intelligent inspection of distribution lines and transmission lines.This paper mainly conducts the following research:Aiming at the problems of complex background of power equipment and defects in distribution line inspection images,large differences in equipment defect sizes,and variable target morphology,this paper proposes an improved YOLOx-based multi-equipment detection and defect recognition method for distribution lines.Based on the YOLOx algorithm,the Receptive Field Block(RFB)is added after the shallow feature layer of the backbone network to increase the sensory field of the network;the Coord Attention(CA)module is added to better obtain the spatial directional feature information of the target and improve the target localization accuracy;the Path Aggregation Network(PAN)is added to improve the target localization accuracy.After the first fusion of features in Path Aggregation Network(PANet),Adaptively Spatial Feature Fusion(ASFF)module is added to achieve efficient re-fusion of multi-scale deep and shallow feature by assigning adaptive weight parameters to features of different scales.In addition,the loss function Bce Loss in YOLOx is replaced by Focal Loss to alleviate the difficult problem of model convergence caused by the imbalance between positive and negative samples of small targets.Experiments are conducted on the self-built distribution dataset of the State Grid Electric Power Research Institute project.The results show that the method proposed in this paper significantly outperforms other comparative methods in terms of performance,and effectively improves the effectiveness of multi-device detection and defect identification in distribution lines.A lightweight network model based on YOLOx is proposed for the problems of limited resources of transmission line UAV inspection platform,high complexity of target detection algorithm and slow inference speed.First,the lightweight Shuffle Net V2＿Plus network is used as the backbone network for feature extraction,and the Depthwise Convolution(DWConv)in the Shuffle Net V2 network is expanded by replacing 3 × 3DWConv in the Shuffle Unit module with 5 × 5DWConv in the Shuffle Unit module,and prune the convolution layer of the model,and prune the 1×1Pointwise Convolution(PWConv)in the Shuffle Unit basic unit module to reduce the network parameters while increasing the network perceptual field.At the same time,add ECA(Efficient Channel Attention)moduleis added in the neck feature fusion part to make the network better focus on important regions and improve the target detection accuracy at a small computational cost.Finally,the ordinary convolution in the YOLOx detection decoupling head is replaced with Depthwise Separable Convolution(DSConv)to further reduce the model complexity.The results show that the inference time of the lightweight network model proposed in this paper is only5.8ms,the model parameters are only 4.361 MB,and the FLOPs are only 10.725 G,and the detection accuracy is high on the combined self-built transmission line dataset.