Bolt Defects Detection For Aerial Transmission Lines From Faster R-CNN With Embedded Dual Attention Mechanism

There are many fittings and bolts on the transmission line,due to they are exposed to the wild all the year round,they are prone to damage,tilt and missing bolt parts.At the same time,the current manual inspection method cannot complete the increasing inspection tasks.The use of deep learning technology can quickly and accurately detect the fitting target on the aerial transmission line image,then cut out the detection results of the fittings from the original image to further detect the defects of the bolts on the fittings.In view of the existing fitting detection methods,it only focuses on a single region,and the connection problem between the fittings has never been considered.From practical experience,various types of fittings have certain regular knowledge when constructing a complete and intact transmission line.Based on this,this paper proposes a knowledge-guided aerial transmission line fittings detection model,which contains two modules to learn the spatial layout and pair-wise relationship of fittings.The first module is an implicit module,starting from the spatial layout of the fittings on the image and taking the relative geometric feature of the region proposal as the input to model the spatial position of the fittings on the image.The second module is an explicit module.By introducing the prior knowledge of the fittings expressed in co-occurrence mode and using the region proposal as the node while the prior knowledge as the edges,a prior knowledge graph constructed.The information spreads on the prior knowledge graph via a gated graph neural network,which realizes a combination of the prior knowledge of the fitting with the features of the region proposal.Test the data set of typical fittings for aerial transmission lines.The experimental results prove that the AP value increased by 4.4% when Io U = {0.5:0.95}and the AP value by 3.9% when Io U = 0.5 compared with those of Faster R-CNN classification.Aiming at the problems that the bolt target in the aerial transmission line image is too small,the difference between different bolt defect categories is small,and the loss of gradient information and the detection effect of bolt areas is not good,this paper proposes a Faster R-CNN aerial transmission line bolt defect detection model embedded with a dual attention mechanism.The model uses the attention mechanism to analyze and enhance the visual features of different scales and positions.Firstly,for the feature of different scales,the network extracts the feature map of each layer and obtain the corresponding attention map,calculates the difference of the attention map for adjacent layers,and then adds it to the loss function as a regularization term to enhance the fine features of the bolt area.Secondly,for the features of different positions,we used the features map to calculate the spatial attention map of the image.Each element in the attention map indicates the degree of similarity between two spatial locations,and then it is used to combine features of each position with the global feature.So,the difference between the bolt and the background is increased.Test on a typical bolt data set of aerial transmission lines.Compared with Faster R-CNN,the model’s m AP value increased by 2.21%,of which the normal bolts,the pin missing bolts and the nut missing bolts increased by 0.29%,5.23% and 1.1%.