| 引用本文: | 万吉林,吴国强,管敏渊,等.基于RetinaNet的变压器图像小部件智能识别方法[J].电力系统保护与控制,2021,49(12):166-173.[点击复制] |
| WAN Jilin,WU Guoqiang,GUAN Minyuan ,et al.Intelligent recognition method for transformer small components based on RetinaNet[J].Power System Protection and Control,2021,49(12):166-173[点击复制] |
|
| 本文已被：浏览 5265次 下载 1544次 |
码上扫一扫！ |
| 基于RetinaNet的变压器图像小部件智能识别方法 |
| 万吉林1,吴国强2,管敏渊2,吴凯2,3,高奥2,施康明2,3,王慧芳1 |
|
|
| (1.浙江大学电气工程学院，浙江 杭州 310027;2.国网浙江省电力有限公司湖州供电公司，
浙江 湖州 313000;3.湖州电力设计院有限公司，浙江 湖州 313000) |
|
| 摘要: |
| 变电站巡检图像中变压器小部件的自动识别是利用变压器图像进行变压器外观异常缺陷识别的基础。为了提高变压器小部件的识别准确性，提出了一种基于RetinaNet的变压器小部件识别方法。首先，对图像目标检测网络RetinaNet进行改进，加入分辨率更高的融合特征图，以解决变压器小部件包含的像素信息过少的问题。然后，提出一种基于位置关联性的变压器小部件概率修正方法，利用识别难度相对较小的大部件位置与相应的小部件之间的位置关联信息，对小部件检测框的预测概率进行修正，以避免其他外形相似部件对目标部件识别的干扰。最后，通过实际变电站巡检图像对变压器小部件识别方法进行实验验证。结果表明，所提出的变压器小部件识别方法在变压器三类小部件的识别准确率以及整体识别准确率上，都具有比较显著的优势。 |
| 关键词: 目标检测 变压器 巡检图像 部件识别 RetinaNet |
| DOI：DOI: 10.19783/j.cnki.pspc.201012 |
| 投稿时间：2020-08-18修订日期：2020-10-25 |
| 基金项目:国家青年科学基金项目资助(62001416)；国网浙江省电力有限公司科技项目资助(2019-HUZJTKJ-09) |
|
| Intelligent recognition method for transformer small components based on RetinaNet |
| WAN Jilin1,WU Guoqiang2,GUAN Minyuan 2,WU Kai2,3,GAO Ao2,SHI Kangming2,3,WANG Huifang1 |
| (1. College of Electrical Engineering, Zhejiang University, Hangzhou 310027, China; 2. Huzhou Power Supply Company,
State Grid Zhejiang Electric Power Company Limited, Huzhou 313000, China; 3. Huzhou Electric Power
Design Institute Company Limited, Huzhou 313000, China) |
| Abstract: |
| Automatic recognition of transformer small components in substation inspection images is the basis of recognizing transformer exterior defects by transformer images. To improve the recognition accuracy, a recognition method based on RetinaNet is proposed. First, the object detection network RetinaNet is improved, and the fusion feature map with higher resolution is added to solve the problem that transformer small components contain too little pixel information. Then, a probability correction method for the components based on location correlation is proposed. This can modify the probability of small component detection boxes according to the location correlation between large components and corresponding small components. This avoids the interference of other similar components to the target component recognition. Finally, the proposed recognition method is verified by the actual substation inspection images, and the result shows that the method has significant advantages both in the recognition accuracy of three types of transformer small components and the overall recognition accuracy.
This work is supported by the National Youth Science Foundation of China (No. 62001416) and the Science and Technology Project of State Grid Zhejiang Electric Power Co., Ltd. (No. 2019-HUZJTKJ-09). |
| Key words: object detection transformer inspection image component recognition RetinaNet |
