DP-YOLOv5: Computer Vision-Based Risk Behavior Detection in Power Grids
| Autor: | Lei Li, Yubo Zheng, Zhe Wang, Xikang Jiang, Lin Zhang, Zheng Yuan, Xinhang Li |
|---|---|
| Rok vydání: | 2021 |
| Předmět: | |
| Zdroj: | Pattern Recognition and Computer Vision ISBN: 9783030880033 PRCV (1) |
| DOI: | 10.1007/978-3-030-88004-0_26 |
| Popis: | The level of safety in power grid construction has been improved by Computer Vision (CV) recently with deep Convolutional Neural Networks (CNN). However, due to environmental complexity and risk behaviors diversity, the current detection algorithms still have false and missing detection problems. This paper quantitatively analyses these practical problems and proposes a Double Precise YOLOv5 (DP-YOLOv5) method. Compared to other state-of-art detectors, DP-YOLOv5 highlights three points: integrating multi objects for each classification to avoid false detection in complex environments, adding standard operation samples for guidance to reduce the missing detection caused by risk behaviors diversity, and using Depthwise Separable convolutional networks to reduce model parameters. The proposed DP-YOLOv5 method is evaluated on a dataset with 2.5k images generated in real power grid operation environments provided by State Grid Jiangsu Electric Power Co., Ltd. Compared with the state-of-art YOLOv5s detector. Experimental results show that the precision of DP-YOLOv5 is 7.1% higher, while the model size is 20% less. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|