Research On Transmission Line Components And Defect Identification Methods Based On Adversarial Continual Learning

The training of transmission line components and defect detection model based on deep learning needs a large number of data sets,and the training consumes extremely computing resources and time resources.However,there are many components on the transmission line,and the defect types of each component are complex.It is very difficult to collect all the data at one time and use it for training.After each new data is collected,it will be fused with the old data to retrain the model,which will lead to a great waste of resources.In this paper,the antagon istic continuous learning technology is introduced into the transmission line component and defect detection task,and the component detection model is cascaded with the defect classification model.Make the detection model obtain the ability to learn new tasks to be detected continuously,and weaken the impact of catastrophic forgetting,so as to save a lot of time and computing resources.Due to its particularity,aerial images of transmission lines have the problems of complex and changeable background and small proportion of detection targets.And some of the images belong to the difficult samples with poor visual information such as shadow and blur.From the perspective of feature fusion,this paper first uses channel attention to make the model pay more attention to the key feature extraction region in complex background.Then,based on ASFF adaptive feature fusion mechanism,the shallow and deep feature maps are fused more reasonably.Finally,the loss function of the detection model is improved to sol ve the problem that the loss function can not accurately reflect the coincidence degree between the real frame and the prediction frame.The experimental results show that compared with the original YOLOx-S,the map value of five types of parts is increased by 1.27%.At present,the defect classification model of transmission line components can not solve the problem of infinite data flow in reality.In this paper,a transmission line component and its defect classification method based on antagonistic cont inuous learning is designed.Firstly,this paper introduces continuous learning technology into transmission line component defect classification task.In order to solve the problem that the classification accuracy is low due to the small difference betwee n the task classes to be classified.The attention mechanism is integrated into the sustainable learning model to enhance the ability of the model to extract subtle features,so as to improve the classification accuracy.Aiming at the unknowability of rank ing in continuous learning tasks,a ranking method based on the number,dispersion and difficulty of data sets is proposed.Multiple groups of control experiments were designed to realize the optimal utilization of continuous learning classification model.Experimental results show that the proposed method significantly improves the classification performance and verifies the effectiveness of this method.