%0 Journal Article
%T 基于改进SSD算法的输电线异物附着故障检测识别技术研究<br>Research on Detection and Recognition Technology of Foreign Object Adhesion Fault in Transmission Line Based on Improved SSD Algorithm
%A 邓天华
%A 赵曙光
%A 刘西钉
%J Smart Grid
%P 36-42
%@ 2161-8771
%D 2022
%I Hans Publishing
%R 10.12677/SG.2022.122005
%X 近年来，我国电力系统紧跟科技前进步伐，得到了空前发展。因而为保证输电系统的稳定运行，对输电线异物附着故障的检测识别方法成为电力行业相关人员的研究热点。为了对输电线异物附着故障进行有效地识别检测，从而提高电力巡检效率，本文结合输电线异物附着故障图像特点，对常见SSD算法(Single Shot MultiBox Detector)进行有效改进，将VGG16特征提取网络替换为ResNet50，并针对原模型在小目标检测中的不足，设计并应用了特征融合模块，且对1241张输电线异物附着故障图像进行数据扩充并制作成包含5000余张图像的数据集，由此训练出目标检测网络模型，最后训练数据集的均值精度mAP (mean Average Precision)在97%左右，达到了故障检测准确性的要求。<br />
In recent years, China’s power system has followed the pace of science and technology and achieved unprecedented development. Therefore, in order to ensure the stable operation of the transmission system, the detection and identification method of foreign matter adhesion fault of transmission line has become a research hotspot of relevant personnel in the power industry. In order to effectively identify and detect the foreign object attachment fault of transmission line, so as to improve the efficiency of power inspection, combined with the image characteristics of foreign object attachment fault of transmission line, this paper effectively improves the common SSD algorithm (Single Shot MultiBox Detector), replaces the VGG16 feature extraction network with ResNet50, and aims at the shortcomings of the original model in small target detection, the feature fusion module is designed and applied, and the data of 1241 foreign object attachment fault images of transmission line are expanded and made into a data set containing more than 5000 images, so as to train the target detection network model. Finally, the mean accuracy mAP (mean Average Precision) of the training data set is about 97%, which meets the requirements of fault detection accuracy.
%K 异物附着，SSD算法，ResNet50，特征融合<br>Foreign Body Attachment
%K SSD Algorithm
%K ResNet50
%K Feature Fusion
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=50802