Research On Electrical Components Surface Defect Detection Of Transmission-Line Based On YOLOX

As an important medium of power transmission,the safe and stable operation of power transmission line is an important guarantee of power supply.In the components of transmission lines,components such as insulators and fittings assume the important role of support,connection and electrical isolation.Therefore,they are also faced with the influence of large voltage and traction force,and are prone to damage,deformation,rust and fall off,which seriously threatens the normal operation of transmission lines.To this end,the power grid company has set up a special power inspection department to troubleshose the transmission line components and replace the faulty components in time.With the gradual maturity of UAV aerial photography technology,the use of UAV instead of manual inspection is gradually becoming the mainstream of current inspection.However,the data collected by UAV is undoubtedly massive,and the manual screening method is difficult to meet the requirements of accurate and efficient detection.In recent years,object detection technology based on computer vision has made remarkable achievements.Introducing it into power inspection can not only greatly improve the efficiency of inspection,but also improve the quality of power inspection.This paper aims to study the main problems of UAV aerial images: poor quality of aerial images,complex and diverse backgrounds,serious occlusion of targets,and an unbalanced number of positive and negative samples.Firstly,aiming at the problem of poor-quality aerial images,the aerial images are manually screened,the images with noise and strong contrast are screened out,and the data is preprocessed.Gaussian filtering was used to denoise the noise generated in image acquisition.Adaptive histogram equalization is used to improve the clarity of the image.Secondly,aiming at the complex and diverse background of images,a cascaded feature enhancement module is designed,which focuses on the important features of the target from the channel dimension and the key regions of the target from the spatial dimension,so as to suppress the influence of irrelevant background information on detection and let the model focus on the local image with important information.Thirdly,aiming at the serious problem of target occlusion,this paper designs a multi-branch attention mechanism,which extracts the features of the target through group convolution,and then sends the features into the channel attention and spatial attention respectively.At the same time,the fusion between the attention is performed to obtain more complementary attention weights,and the target detection is realized from multiple angles to solve the problem of serious target occlusion.Lastly,for the problem of an unbalanced number of positive and negative samples,this paper introduces the Focal loss confidence loss function,adjusts the learning weights of positive and negative samples,and optimizes the gradient descent direction,to effectively reduce the influence of unbalanced samples on the detection effect.In power inspection,it is a general trend to replace manual inspection by UAV inspection.Therefore,this paper is oriented to the defect detection of transmission line components such as insulators and fittings,through the study of computer vision technology based on deep learning,effective image analysis is realized on the basis of existing software tools.The proposed method is not only suitable for surface defect detection of transmission line components,but also can be extended to other inspection tasks of transmission lines to improve the efficiency of inspection.