Research On The Power Equipment Identification Technology Of High-voltage Transmission Line Based On Deep Learning

Line inspection is an important guarantee for the smooth operation of the power supply system.With the rapid development of deep learning,the target recognition algorithm based on deep learning has become an important part of the field of target recognition,aiming to use the neural network to judge and process the corresponding images and identify the objects in the images.At present,it is widely used in electric power equipment identification,defect detection and other fields.At present,the common kinds of power equipment identification algorithms have achieved good results in the inspection of high voltage transmission lines,but there are still problems in using the identification technology of high voltage transmission lines based on deep learning in the actual line inspection.This article is for inspection The following contents of the identification technology of high voltage transmission lines based on deep learning are studied on the problems of the illumination of the equipment,the scale of the power equipment,the scale transformation of the power equipment and the camera.1.In the actual inspection task,there are large lighting differences in the image background of power equipment,resulting in invalid recognition.A Retinex image preprocessing model based on grayscale GSA was constructed,and the preprocessing operation was performed before the image was input into a deep learning algorithm.The image are grayscale information by using the GSA optimization algorithm Strong,to get a more normal gray scale image.The image was transformed from RGB to HSV space,the image contrast was enhanced based on Retinex algorithm,and the illumination layer L separated from the image lightness V was shaded.Ensure that the image bright part is not exposed,effectively save the bright part image feature information,improve the image dark part feature information,and effectively achieve the balance of light and shade of the image.2.An improved Faster RCNN algorithm is constructed to solve the difficulty of identifying remote power equipment.By comparing the recognition effect of different feature extraction network,the Res Net101 skeleton network is selected For the feature extraction network,the algorithm improves the ability of the algorithm to extract image feature information,and the deconvolution layer is added at the end of the shared convolution layer,which has the problem of low recognition accuracy.In the RPN module,the target candidate box is more suitable for the power device scale,so that the candidate box is more suitable for the power device in the image.Thus,the availability of the identification algorithm can effectively identify all kinds of power equipment in the inspection process.3.The large difference in width and height of power equipment,resulting in missing inspection of conventional anchor frame.The assignment of Anchor box is adjusted so that the recognition algorithm improves the recognition energy of extreme scale targets Force,improve the algorithm identification effect.In terms of the algorithm loss function,compared with the traditional method of calculating the distance between the sample and the real box directly,the Euclidean distance calculation of the central point is used to obtain the intersection area relationship between the two.The improved loss function can effectively ensure the regression of the target box and avoid the occurrence of non-convergence.Finally,the image enhancement algorithm and the improved Faster RCNN algorithm are combined to construct a more suitable system for power equipment target identification,which realizes the effectiveness of power equipment identification for high-voltage transmission line.