| 基于改进 YOLOv4 的嵌入式变电站仪表检测算法 |
| |
| 作者姓名： | 范新南 黄伟盛 史朋飞 辛元雪 朱凤婷 周润康 |
| |
| 作者单位： | 1. 江苏省输配电装备技术重点实验室，江苏 常州 213022；
2. 河海大学物联网工程学院，江苏 常州 213022 |
| |
| 基金项目： | 江苏省输配电装备技术重点实验室开放研究基金项目(2021JSSPD03) |
| |
| 摘 要： | 随着机器人技术的快速发展，智能机器人广泛应用于变电站巡检，针对目前目标检测算法参数量过大且嵌入式设备性能有限，难以在嵌入式平台上实现实时检测的问题，提出了一种基于改进 YOLOv4 的嵌入式变电站仪表检测算法。以 YOLOv4 为基础，采用MobileNetV3 作为主干特征提取网络，在保证模型能够有效提取特征的情况下，降低运算量，提高检测速度；与此同时，将特征提取后的路径聚合网络(PANet)中的卷积运算替换成深度可分离卷积；采用迁移学习的训练策略克服模型训练困难问题；最后，利用TensorRT对改进后的模型进行重构和优化，实现快速和高效的部署推理。改进后的算法在嵌入式端 NVIDIA Jetson Nano上进行了测试，实验结果表明，在牺牲了较少精度的情况下，检测速度提高了 2 倍，达到 15 FPS，为边缘计算场景下的仪表实时检测提供了可能。
|
| 关 键 词： | 深度学习 变电站仪表 目标检测 YOLOv4 迁移学习 |
Embedded substation instrument detection algorithm based onimproved YOLOv4 |
| |
| Authors: | FAN Xin-nan HUANG Wei-sheng SHI Peng-fei XIN Yuan-xue ZHU Feng-ting ZHOU Run-kang |
| |
| Affiliation: | 1. Jiangsu Key Laboratory of Power Transmission & Distribution Equipment Technology, Changzhou Jiangsu 213022, China;
2. College of Internet of Things Engineering, Hohai University, Changzhou Jiangsu 213022, China |
| |
| Abstract: | With the rapid development of robotics technology, intelligent robots are widely used in substation inspections. Aiming at the problem that the current target detection algorithms have too many parameters and the performance of embedded devices is limited. It is difficult to achieve real-time detection on the embedded platform. A n improved YOLOv4 embedded substation instrument detection algorithm is proposed. The algorithm is based on YOLOv4 and uses MobileNetV3 as the backbone feature extraction network. It reduces the amount of calculation and increases the detection speed while ensuring that the model can effectively extract features. At the same time, the convolution operation in the path aggregation network (PANet) is replaced with a depthwise separable convolution after feature extraction; the training strategy of transfer learning is used to overcome the difficult problem of model training. Finally, the improved model is optimized by TensorRT to achieve fast and efficient deployment reasoning. The improved algorithm is tested on the embedded NVIDIA Jetson Nano, and the experimental results show that the detection speed is increased by 2 times to 15 FPS at the expense of less accuracy. This provides the possibility for real-time instrument detection in edge computing scenarios. |
| |
| Keywords: | |
|
| 点击此处可从《》浏览原始摘要信息 |
|
点击此处可从《》下载全文 |
