摘要
输电线路绝缘子故障影响电力系统供电可靠性,为了实现航拍图像中绝缘子的准确检测,本研究提出了一种基于改进Faster-RCNN网络的输电线路航拍绝缘子目标检测方法(ScSGB-RCNN),主要工作有:1)针对检测算法精度低的问题,采用自校准卷积结构(Self-calibrated convolutional Network,ScNet)和ConvNeXt网络构建了ScConvNeXt主干网络,通过融合多个卷积注意力模块,扩大网络的全局感受野,提升检测精度。2)为优化不同尺度绝缘子目标的特征提取能力,提出一种轻量化的特征金字塔结构SFPN,融入到ScConvNeXt网络中,降低计算参数量。3)为提高模型收敛速度和检测精度,采用GeLU激活函数改进FRN(Filter Response Normalization,FRN)归一化函数,提升网络的非线性输出能力。4)设计了BIoU并重新构建定位损失函数。实验结果表明,本研究提出的方法较原算法精度提高22.4%,模型收敛速度提升4倍,FPS提高8.7帧/秒,优于Faster-RCNN、SSD、YOLOv3、YOLOv5、YOLOv7等算法。该方法可为输电线路航拍绝缘子检测提供技术参考。
Defective insulators on transmission lines affect the reliability of power supply systems.In order to achieve accurate detection of insulators in aerial images,the author proposes a transmission lines insulators image target detection method(ScSGB-RCNN)based on improved Faster-RCNN network.The main tasks are as follows:(1)To solve the problem of low accuracy of the detection algorithm,Self-calibrated convolutional Network(ScNet)and ConvNeXt network are used to construct the ScConvNeXt backbone network.By integrating multiple convolutional attention modules,enlarge the global receptive field of the network and improve the detection accuracy.(2)In order to optimize the feature extraction capability of different scale insulator targets,a lightweight feature pyramid structure SFPN was proposed and integrated into ScConvNeXt network to reduce the number of calculation parameters.(3)In order to improve the convergence speed and detection accuracy of the model,GeLU activation function is used to improve the Filter Response Normalization(FRN)function and improve the nonlinear output capability of the network.(4)BloU is designed and the positioning loss function is reconstructed.Experimental results show that compared with the original algorithm,the accuracy of the proposed method is improved by 22.4%,the model convergence speed is increased by 4 times,and the FPS is increased by 8.7 frames/second,which is superior to Faster-RCNN,SSD,YOLOv3,YOLOv5,YOLOv7 and other algorithms.This method can provide technical reference for the aerial photo insulator detection of transmission lines.
作者
曾业战
段志超
郭彦东
钟春良
ZENG Yezhan;DUAN Zhichao;GUO Yandong;ZHONG Chunliang(College of Electrical and Information Engineering,Hunan University of Technology,Zhuzhou 412007,China)
出处
《电瓷避雷器》
CAS
2024年第1期161-169,共9页
Insulators and Surge Arresters
基金
湖南省自然科学基金资助项目(编号:2020JJ4276)。
关键词
输电线路
绝缘子
深度学习
目标检测
transmission lines
insulators
deep learning
target detection