| 引用本文: | 魏贤哲,卢 武,赵文彬,等.基于改进Mask R-CNN的输电线路防外破目标检测方法研究[J].电力系统保护与控制,2021,49(23):155-162.[点击复制] |
| WEI Xianzhe,LU Wu ZHAO,Wenbin,et al.Target detection method for external damage of a transmission line based on an improved Mask R-CNN algorithm[J].Power System Protection and Control,2021,49(23):155-162[点击复制] |
|
| 摘要: |
| 随着现代电力系统的不断发展，电网规模越来越大，外破原因引发的故障已经成为架空输电线路故障的主要原因之一。在架空输电线路视频监控中，使用传统的边界框式目标检测方法进行外破预警时，误报或漏报的情况时有发生。掩模实例分割神经网络(Mask-RCNN)训练时使用的像素级掩模标注数据集成本较高，限制了该算法的大规模应用。针对这些问题，将改进的Mask-RCNN网络应用到输电线路外破目标检测领域，在数据集标注过程中，使用边界框标注代替部分掩模标注。训练时，将检测分支的特征迁移到掩模分支。实验结果表明，改进后的算法能够在掩模标注样本占比80%的条件下，对常见外破类别的平均识别准确率高于91%，为输电线路外破隐患的准确识别与分割提供了一种可行的思路。 |
| 关键词: 输电线路 外力破坏 Mask R-CNN 迁移学习 数据增广 |
| DOI：DOI: 10.19783/j.cnki.pspc.210482 |
| 投稿时间：2021-04-24修订日期：2021-07-30 |
| 基金项目:国家自然科学基金项目资助(61502297) |
|
| Target detection method for external damage of a transmission line based on an improved Mask R-CNN algorithm |
| WEI Xianzhe,LU Wu ZHAO,Wenbin,WANG Daolei |
| (1. College of Electrical Engineering, Shanghai University of Electric Power, Shanghai 200090, China;
2. College of Energy and Mechanical Engineering, Shanghai University of Electric Power, Shanghai 200090, China) |
| Abstract: |
| With the continuous development of modern power systems and the increasing scale of power grids, faults caused by external damage have become one of the main causes of overhead transmission line failures. In overhead transmission line video surveillance, false or missed alarms occur when using the traditional bounding box target detection method for external damage warning. The high cost of integrating the pixel-level mask annotation data used in the training of the mask instance segmentation neural network (Mask-RCNN) limits the large-scale application of this algorithm. To address these problems, this paper applies an improved Mask-RCNN network to the field of transmission line external breakage target detection by using bounding box annotation instead of partial mask annotation in the process of dataset annotation. During training, the features of the detection branch are migrated to the mask branch. The experimental results show that the improved algorithm can achieve an average recognition accuracy of more than 91% for common external damage categories with 80% of the mask labeled samples, providing a feasible idea for accurate identification and segmentation of external damage hazards on transmission lines.
This work is supported by the National Natural Science Foundation of China (No. 61502297). |
| Key words: transmission line external force damage Mask R-CNN transfer learning data augmentation |
