| 引用本文: | 杨剑锋,秦 钟,庞小龙,贺志华,崔春晖.基于深度学习网络的输电线路异物入侵监测和识别方法[J].电力系统保护与控制,2021,49(4):37-44.[点击复制] |
| YANG Jianfeng,QIN Zhong,PANGXiaolong,HE Zhihua,CUI Chunhui.Foreign body intrusion monitoring and recognition method based on Dense-YOLOv3 deep learning network[J].Power System Protection and Control,2021,49(4):37-44[点击复制] |
|
| 摘要: |
| 为解决输电线路异物入侵在线监测图像样本量较小的问题，针对异物图像特点，提出了一种基于深度学习的输电线路异物入侵监测和识别方法。首先选取典型正常运行输电线路图像和目标异物图像，采用条件生成对抗网络算法对有异物入侵的输电线路图像进行样本扩充。然后将Dense-net网络替代YOLOv3网络中倒数第二层网络，建立Dense-YOLOv3深度学习网络模型。使用实际图像样本和扩充图像样本数据训练和测试深度学习网络，实现输电线路异物入侵监测和识别。该型深度学习网络算法可以对风筝、鸟巢、垃圾、机械施工类异物入侵情况进行有效识别，识别准确率分别达到98%、96%、90%和100%。 |
| 关键词: 输电线路 异物入侵 深度学习 生成对抗网络 YOLOv3网络 Dense-net网络 |
| DOI：DOI: 10.19783/j.cnki.pspc.200508 |
| 投稿时间：2020-05-06修订日期：2020-06-19 |
| 基金项目:国家电网公司科技项目资助(SGITG-2018ZXCG-FF) |
|
| Foreign body intrusion monitoring and recognition method based on Dense-YOLOv3 deep learning network |
| YANG Jianfeng,QIN Zhong,PANGXiaolong,HE Zhihua,CUI Chunhui |
| (1. State Grid Ningxia Electric Power Co., Ltd., Yinchuan 750001, China; 2. State Grid Yinchuan Power Supply Company,
Yinchuan 750001, China; 3. State Grid Ningxia Maintenance Company, Yinchuan 750001, China;
4. Nanjing Youkuo Electric Technology Co., Ltd., Nanjing 211100, China) |
| Abstract: |
| The small sample size of transmission line foreign body invasion online monitoring imagea, and of the image features of foreign body present a problem. Thus a deep learning method for foreign body invasion monitoring and recognition of transmission lines is proposed. First, the images of typical transmission lines and foreign bodies in normal operation are selected, and the images of transmission lines invaded by foreign bodies are expanded by means of a conditional generation antagonism network algorithm. Then the Dense-net network is replaced by the penultimate network in the YOLOv3 network, and the deep learning network model of Dense-YOLOv3 is established. A real image sample and the extended image sample data are used to train and test the deep learning network to realize the foreign body intrusion monitoring and recognize the transmission line. This deep learning network can effectively identify foreign objects such as kite, model aircraft, garbage and mechanical construction, with a recognition accuracy reaching 98%, 96%, 90% and 100%, respectively.
This work is supported by the Science and Technology Project of State Grid Corporation of China (No. SGITG- 2018ZXCG-FF). |
| Key words: transmission line foreign body intrusion deep learning generative antagonistic network YOLOv3 network Dense-net network |
