摘要
为了精准地识别无人机巡检图形中的小目标绝缘子及缺陷,本文提出了一种基于改进的深度学习目标检测网络(YOLOv4)的输电线路绝缘子缺陷检测方法。首先,通过无人机航拍及数据增强获得足够的绝缘子图像,构造绝缘子数据集。其次,利用绝缘子图像数据集训练YOLOv4网络,在训练过程中采用多阶段迁移学习策略和余弦退火学习率衰减法提高网络的训练速度和整体性能。最后,在测试过程中,对存在小目标的图像采用超分辨率生成网络,生成高质量的图像后再进行测试,以提高识别小目标的能力。实验结果表明,与Faster R-CNN和YOLOv3相比,所提算法在平均分类精度和每帧检测速率方面均有较大提升,性能表现优异。
In order to accurately identify small target insulators and defects in UAV inspection graphics,a defect detection method for transmission line insulators based on improved deep learning target detection network namely YOLOv4 was proposed.First,sufficient insulator images were collected by the means of drone and data enhancement to construct insulator dataset.Secondly,insulator image data set was used to train YOLOv4 network.During the training process,multi-stage transfer learning strategy and cosine annealing attenuation learning rate method were adopted to improve the training speed and overall performance of the network.Finally,in the test process,the super-resolution reconstruction generative adversarial network was introduced to generate high-quality images for low-confidence images,and then the test was carried out again to improve identification ability of small objects.The experimental results show that,compared with Faster R-CNN and YOLOv3,the average classification accuracy and detection rate per frame of the proposed algorithm are greatly improved and the performance is excellent.
作者
高伟
周宸
郭谋发
GAO Wei;ZHOU Chen;GUO Mou-fa(College of Electrical Engineering and Automation,Fuzhou University,Fuzhou 350108,China)
出处
《电机与控制学报》
EI
CSCD
北大核心
2021年第11期93-104,共12页
Electric Machines and Control
基金
国家自然科学基金(51677030)
晋江市福大科教园区发展中心科研项目(2019-JJFDKY-23)。
关键词
绝缘子
缺陷检测
YOLOv4
数据增强
多阶段迁移学习
超分辨率生成网络
insulator
defect detection
YOLOv4
data enhancement
multi-stage transfer learning
super-resolution reconstruction generative adversarial network(SR-GAN)
