Research On Defect Detection In Power Transmission Lines Based On Deep Learning

With the development of the power system,defect detection of transmission lines has become particularly important.This paper takes the detection of insulator defects in transmission lines as an example to test the effectiveness of the model in defect detection of transmission lines.Insulators are an important component used in power systems to maintain insulation between electrical equipment and play a role in isolating conductors and protecting equipment.However,due to environmental factors and long-term use,insulators have various defect problems such as cracks,dirt,peeling,etc.,and other components of transmission lines are in the same environment as insulators and will also have different defects.These defect problems will have adverse effects on the stable operation of the power system.Therefore,how to quickly and accurately detect defects in various components of transmission lines has become an important research topic in the current field of power systems.Currently,traditional defect detection methods for transmission lines mainly rely on manual visual inspection or the use of professional testing equipment.This method has problems such as low defect detection efficiency and low accuracy.With the development of computer vision and deep learning,image defect detection methods based on deep learning have gradually been introduced into transmission line defect detection.Deep learning has superior performance in image processing,especially in object detection.Algorithms based on deep learning can achieve high-speed and accurate defect detection.This paper proposes an improved model,INS-YOLOX,based on a one-stage detector,effectively solving the accuracy and speed problems in the detection of defects in transmission lines.The paper mainly tests the improved model on the detection of insulator defects in transmission lines.Insulator defects have a complex background and various targets,with little difference between defective and non-defective insulator images,and the targets are small.In order to improve the detection accuracy,a new loss function,L-SIoU,is proposed in this paper.It not only considers the angle between the required regression vectors based on SIoU but also redefines the punishment index,and considers the impact of the aspect ratio of the predicted box and regression box on the results,improving the accuracy and speed of inference.Then,Varifocal Loss is used to replace BCE_Loss,improving the accuracy of the model in detecting imbalanced datasets.Finally,the CBAM attention mechanism is added to the Decoupled Head structure to enhance the feature point extraction of the image.INSYOLOX also introduces the SGD optimizer and effectively optimizes the model’s training parameters.In the experiment,this paper tests the INSYOLOX model on 6227 images as the dataset and verifies its effectiveness.The test results show that INS-YOLOX can achieve real-time detection,with a final map of 83.2,an increase of 11.1%in accuracy over the baseline model of YOLOX-tiny and 6%higher accuracy than the latest Yolov7.This achievement has significant practical value and application prospects.