基于深度学习的输电线路绝缘子缺陷识别研究

Research on Insulator Defect Recognition of Transmission Line Based on Deep Learning

输电线路中绝缘子的缺陷识别检测为维持电力系统的安全稳定性十分重要,但由于绝缘子所处背景的复杂性及缺陷在航拍图像中占比较小的问题,使得绝缘子检测及其缺陷识别具有一定的挑战。针对上述,文章提出了一种基于深度学习的输电线路绝缘子缺陷识别研究。通过全局检测和局部分割的两级检测方法对绝缘子进行缺陷识别。首先全局检测采用YOLOv3算法对绝缘子进行定位并切割出仅包含绝缘子的图像,扩大缺陷部分在整个图像上的占比,然后将切割后的图像送入U-Net网络中对绝缘子及缺陷进行像素级检测,提高对缺陷识别的精度。通过实验验证分析,两级检测网络对缺陷的识别准确率同只用YOLOv3进行一级检测比提高2.8%,该方法能较为有效地、准确地识别绝缘子缺陷。

In the transmission line insulator defect recognition detection in order to maintain the security of power,system stability is very important.But because of the complexity of the insulator are the background and the defects of smaller problems in aerial images,makes the insulator and its defect identification has a certain challenges.For the above,the article puts forward a transmission line insulator defect recognition based on deep learning.Defect identification of insulators is carried out by global detection and local segmentation.First of all,the YOLOv3 algorithm is adopted for global detection to locate insulators and cut out images containing only insulators to expand the proportion of the defective part in the whole image.Then,the cut images are sent into the U-Net network for pixel-level detection of insulators and defects to improve the accuracy of defect identification.Through experimental verification and analysis,the accuracy of defect identification of the two-stage detection network is 2.8%higher than that of the first-stage detection only using YOLOv3.This method can identify insulator defects more effectively and accurately.