Research On Detection Algorithm Of Transmission Line Components Based On Convolutional Neural Network

Most components of the transmission line,which is the most important part of the transmission system,are exposed to the outdoor environment all the year round.These exposed components are affected not only by the internal factors,but also be damaged by the external environment,so we have to check these components regularly.However,the traditional manual inspection method cannot guarantee the check speed and consumes a lot of resource.And,the helicopter inspection method is expensive.But UAV inspections are fast,accurate,and cost-effective.Therefore,inspecting the developments by the UAV equipped with intelligent detection algorithm is the trend of future transmission line components inspection.Most prior methods focus on the high-resolution input images and components in large size.However,some images in the UAV inspection dataset are low-resolution and some components are small.To mitigate these problems,we propose a dynamic superresolution-based transmission line,bolts and vibration dampers components inspection algorithm,which combines a dynamic super-resolution module and an object detection module.The dynammic super-resolution module can adaptively select a suitable superresolution algorithm for processing the input images according to their features.Then the processed images are sent to the YOLOv5 algorithm to complete the object detection process.Experiments prove the superiority of our proposed method on lowresolution images.And in the detection of insulator and its defects,the large difference between the size of insulator and that of the defect leads to the omission of insulators.Taking this situation into consideration,we modify the YOLOv5 model by incorporating the attention mechanism We choose ResNet-18 as our backbone and add the convolutional block attention module to it.In addition,we add the multi-head attention module to the neck network and optimize the structure and bounding box regression loss functions.Experimental results show that our method mitigates the omission problem of the YOLOv5 and improves the detection performance of insulator and insulator defects.This paper focus on the transmission line components inspection algorithm based on convolutional neural network.The algorithm we proposed realizes the location and identification of transmission line components and defects,which can be applied to UAV inspections to improve the efficiency of transmission line components detection.