Research On Fault Detection Of RESNET Transmission Line Based On Attention Mechanism

The smooth operation of transmission lines is an important support to ensure power transmission.Therefore,regular inspection of transmission lines is the basic task of maintaining grid security.For the more manual inspection methods,the shortcomings of manpower waste and high risks cannot be ignored.With the development of power grid intelligence and deep learning technology,the combination of drone inspection and image processing technology has been integrated into the further research of transmission line inspection.The use of image classification algorithms can effectively realize the intelligence of transmission line fault detection,making the detection safer and more efficient.Through research on detection methods of transmission lines at home and abroad,it is found that the existing image classification algorithms not only have room for improvement in accuracy,but also have huge model parameters,which require high computer performance and are difficult to solve.In the actual transmission line inspection work.Therefore,this paper studies these deficiencies.Firstly,this paper analyzes the common faults of transmission lines,and makes the data set TFDS and its enhanced data set TFDS-PRO according to its image features.Then resnet50 is selected as the basic model of transmission line image classification.Aiming at the disadvantage of large amount of parameters,the convolution layer is improved to obtain DOC_R50 model and tested on cifar common data set and transmission line fault data set TFDS and TFDS-PRO respectively.The experimental results show that DOC_R50 model is not only improved in classification accuracy,but also competitive in parameter quantity.The best accuracy of fault classification in transmission line is 97.68%.After further analysis,it is found that DOC_R50 model network is difficult to effectively extract features when the transmission line is in a complex environment.In order to further improve the network performance,attention mechanism is introduced.Based on the existing attention modules,two new attention modules DEA and TDAM modules are constructed.The improved two modules have the advantages of small parameters and strong feature extraction ability.They are embedded in DOC_R50 network,a model ER50 which can efficiently classify transmission line faults is obtained.Taking TFDS-PRO as the experimental data set,the effectiveness of attention module is verified by ablation experiment,and the extraction details of ER50 network are more intuitively displayed by feature visualization.In addition,in order to realize the engineering significance of the proposed image classification model,a transmission line fault classification system based on this algorithm is designed.Experiments show that the accuracy of transmission line fault classification model ER50 is 99.87%,and the overall parameters remain at a small level,only 0.806G,which provides an experimental basis and reference for intelligent inspection of power grid.Figure[54]table[19]reference[73]...