面向航拍图像的绝缘子弱特征缺陷两阶段检测方法

Two-stage detection method for weak feature defects of insulators in aerial images

绝缘子缺陷是输电线路自动化巡检的重点关注对象。然而,由于绝缘子弱特征缺陷(污秽、裂口、腐蚀和灼烧等)存在特征模糊、类内差异大和类间相似度高等问题,给绝缘子缺陷精准检测带来极大挑战。为此,提出一种面向航拍图像的绝缘子弱特征缺陷两阶段检测方法。首先,针对巡线航拍图像背景复杂、尺度变化和绝缘子遮挡等问题,通过引入空间细化卷积注意力和语义流对齐模块,提出一种改进的YOLOv5框架,实现绝缘子准确定位。然后,采用DeepLabv3+实现被定位绝缘子中缺陷的精准分割和可靠识别。最后,基于真实巡检航拍图像数据集进行实验。结果表明,所提两阶段法在绝缘子识别和弱特征缺陷检测阶段的准确率分别达到了86.70%和77.42%,能较好解决航拍图像中绝缘子缺陷特征弱、目标区域小等难题,为智能化巡检技术发展提供了可借鉴的思路。

Insulator defects are the focus of automatic inspection in power transmission lines.However,weak feature defects,such as pollution,cracks,corrosion,and burn,have fuzzy characteristics,large intra-class differences,and high inter-class similarity,which bring great challenges to the accurate detection of insulator defects.Therefore,a two-stage detection method for weak feature defects of insulators in aerial images is proposed in this paper.Firstly,aiming at the issues such as complex backgrounds,scale variations,and insulator obstructions in aerial images for power line inspection,an improved YOLOv5 framework is proposed by introducing spatially refined convolutional block attention module and flow-based alignment module to achieve accurate localization of insulators.Then,DeepLabv3+is used to achieve accurate segmentation and reliable identification of defects in positioned insulators.Finally,experiments are conducted using a real inspection aerial image dataset.The experimental results show that the proposed two-stage method achieves accuracies of 86.70%and 77.42%in the stages of insulator recognition and weak feature defect detection,respectively.It effectively addresses the challenges of weak defect features and small target areas in aerial images for insulators,providing a reference for the development of intelligent inspection technology.