基于无人机航拍的绝缘子掉串实时检测研究

doi:10.16183/j.cnki.jsjtu.2021.416

摘要/Abstract

摘要：

由无人机代替人工进行电力绝缘子巡检具有重要意义,针对无人机的上位机算力和存储资源有限的问题,提出一种适用于绝缘子掉串故障检测的实时目标检测改进算法.以YOLOv5s检测网络为基础,将颈部结构中路径聚合网络替换为双向特征金字塔网络,以提升特征融合能力;使用DIoU优化损失函数,对模型进行γ系数的通道剪枝和微调,总体上提升检测网络的精度、速度和部署能力;在网络输出处进行图像增强以提升算法可用性.在特殊扩增的绝缘子故障数据集下测试,相较于原始的YOLOv5s算法,改进算法在精度平均值上提升了3.91%,速度提升了25.6%,模型体积下降了59.1%.

关键词: 无人机, 绝缘子掉串, 双向特征金字塔网络结构, γ系数剪枝微调, DIoU损失函数, 图像增强

Abstract:

It is of great significance for unmanned aerial vehicle(UAV) to replace manual inspection of power insulators. Aimed at the problem of limited computing power and storage resources of the UAV, an improved real-time target detection algorithm suitable for insulator drop string failure detection is proposed. Based on the YOLOv5s detection network, first, the PANet networks in neck are replaced with bi-directional feature pyramid network(BiFPN) to improve the feature fusion ability. Next, DIoU is used to optimize the loss function to optimize the model. The channel pruning and fine tuning of the γ coefficient generally improve the accuracy, speed, and deployment ability of the detection network. Finally, the image is enhanced at the network output to improve the availability of the algorithm. The proposed algorithm is tested under a specially expanded insulator fault data set. The results show that compared with the original YOLOv5s algorithm, the average accuracy of the proposed algorithm is improved by 3.91%, the detection speed is improved by 25.6%, and the model volume is reduced by 59.1%.

Key words: unmanned aerial vehicle (UAV), insulator drop string, bi-directional feature pyramid network (BiFPN), γ coefficient pruning fine adjustment, DIoU loss function, image enhancement

中图分类号:

TM855

李登攀, 任晓明, 颜楠楠. 基于无人机航拍的绝缘子掉串实时检测研究[J]. 上海交通大学学报, 2022, 56(8): 994-1003.

LI Dengpan, REN Xiaoming, YAN Nannan. Real-Time Detection of Insulator Drop String Based on UAV Aerial Photography[J]. Journal of Shanghai Jiao Tong University, 2022, 56(8): 994-1003.

图/表 20

图1

图2

图3

图4

图5

图6

图7

图8

图9

图10

图11

图12

图13

图14

图15

表1

所有网络模型的测试结果汇总

组别	网络模型	mAP(50%)			$t -$ /ms	N_RAM/MB	体积/MB
组别	网络模型	绝缘子整体	掉串故障位置	平均值	$t -$ /ms	N_RAM/MB	体积/MB
1	YOLOv5s	0.908	0.798	0.853	12.4	7.02	13.7
2	YOLOv5s+数据集扩充	0.927	0.861	0.894	12.5	7.02	13.7
3	YOLOv5s+数据集扩充+DIoU	0.939	0.862	0.901	12.5	7.02	13.7
4	YOLOv5s+数据集扩充+BiFPN	0.929	0.847	0.888	12.2	7.08	13.9
5	YOLOv5s+数据集扩充+ BiFPN +DIoU	0.945	0.868	0.907	12.7	7.08	13.9
6	YOLOv5s+数据集扩充+ BiFPN + DIoU+γ系数剪枝微调	0.951	0.906	0.929	9.3	2.79	5.60

表1

图16

图17

表2

轻量网络测试数据对比汇总

网络模型	mAP(50%)			$t -$ /ms	N_RAM/ MB	体积/ MB
网络模型	绝缘子整体	掉串故障位置	平均值	$t -$ /ms	N_RAM/ MB	体积/ MB
YOLOv3-tiny	0.901	0.897	0.899	5.4	8.67	16.6
YOLOv5n	0.883	0.636	0.760	9.5	1.76	3.74
YOLOv5s	0.927	0.861	0.894	12.5	7.02	13.7
YOLOXs	0.977	0.909	0.943	12.7	8.94	68.5
本文	0.951	0.906	0.929	9.3	2.79	5.60

表2

图18

参考文献 23

[1]	方挺, 董冲, 胡兴柳, 等. 航拍图像中绝缘子串的轮廓提取和故障检测[J]. 上海交通大学学报, 2013, 47(12): 1818-1822.
	FANG Ting, DONG Chong, HU Xingliu, et al. Contour extraction and fault detection of insulator strings in aerial images[J]. Journal of Shanghai Jiao Tong University, 2013, 47(12): 1818-1822.
[2]	李映国, 杨宏, 徐郁, 等. 基于Alexnet网络的绝缘子自爆无人机巡检技术研究[J]. 智慧电力, 2021, 49(8): 104-109.
	LI Yingguo, YANG Hong, XU Yu, et al. Insulator self-explosive inspection technology based on alexnet network in UAV grid inspection[J]. Smart Power, 2021, 49(8): 104-109.
[3]	王卓, 王玉静, 王庆岩, 等. 基于协同深度学习的二阶段绝缘子故障检测方法[J]. 电工技术学报, 2021, 36(17): 3594-3604.
	WANG Zhuo, WANG Yujing, WANG Qingyan, et al. Two stage insulator fault detection method based on collaborative deep learning[J]. Transactions of China Electrotechnical Society, 2021, 36(17): 3594-3604.
[4]	李彬, 汪诚, 丁相玉, 等. 改进YOLOv4的表面缺陷检测算法[J/OL].(2021-11-15)[2021-12-06]. https://doi.org/10.13700/j.bh.1001-5965.2021.0301.
	LI Bin, WANG Cheng, DING Xiangyu, et al. Surface defect detection algorithm based on improved YOLOv4[J/OL].(2021-11-15)[2021-12-06]. https://doi.org/10.13700/j.bh.1001-5965.2021.0301.
[5]	毛昭勇, 王亦晨, 王鑫, 等. 面向高速公路的车辆视频监控分析系统[J]. 西安电子科技大学学报, 2021, 48(5): 178-189.
	MAO Zhaoyong, WANG Yichen, WANG Xin, et al. Vehicle video surveillance and analysis system for the expressway[J]. Journal of Xidian University, 2021, 48(5): 178-189.
[6]	符惠桐, 王鹏, 李晓艳, 等. 面向移动目标识别的轻量化网络模型[J]. 西安交通大学学报, 2021, 55(7): 124-131.
	FU Huitong, WANG Peng, LI Xiaoyan, et al. Lightweight network model for moving object recognition[J]. Journal of Xi'an Jiaotong University, 2021, 55(7): 124-131.
[7]	JOCHER G. Yolov5[DB/OL]. (2021-10-28)[2021-12-06]. https://github.com/ultralytics/yolov5.
[8]	REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, NV, USA: IEEE, 2016: 779-788.
[9]	REDMON J, FARHADI A. YOLO9000: Better, faster, stronger[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, HI, USA: IEEE, 2017: 6517-6525.
[10]	REDMON J, FARHADI A. YOLOv3: An incremental improvement[EB/OL]. (2018-04-08)[2021-12-06]. https://arxiv.org/abs/1804.02767.
[11]	BOCHKOVSKIY A, WANG C Y, LIAO H Y. YOLOv4: Optimal speed and accuracy of object detection[J/OL]. (2020-04-23)[2021-12-06]. https://arxiv.org/abs/2004.10934v1.
[12]	FACEBOOK AI RESEARCH. PASCAL visual object classes challenge[EB/OL]. (2012-01-01) [2021-08-20]. https://paperswithcode.com/dataset/pascal-voc.
[13]	GOOGLE. Global wheat detection[EB/OL]. (2020-08-05)[2021-08-20]. https://www.kaggle.com/c/global-wheat-detection.
[14]	LIN T Y, MAIRE M, BELONGIE S, et al. Microsoft COCO: Common objects in context[M]// Computer vision-ECCV 2014. Germany: Springer, 2014: 740-755.
[15]	WANG C Y, MARK LIAO H Y, WU Y H, et al. CSPNet: A new backbone that can enhance learning capability of CNN[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. Seattle, WA, USA: IEEE, 2020: 1571-1580.
[16]	REZATOFIGHI H, TSOI N, GWAK J, et al. Generalized intersection over union: A metric and a loss for bounding box regression[C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach, CA, USA: IEEE, 2019: 658-666.
[17]	TAN M X, PANG R M, LE Q V. EfficientDet: Scalable and efficient object detection[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle, WA, USA: IEEE, 2020: 10778-10787.
[18]	ZHENG Z H, WANG P, LIU W, et al. Distance-IoU loss: Faster and better learning for bounding box regression[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7): 12993-13000. doi: 10.1609/aaai.v34i07.6999 URL
[19]	LIU Z, LI J G, SHEN Z Q, et al. Learning efficient convolutional networks through network slimming[C]// 2017 IEEE International Conference on Computer Vision. Venice, Italy: IEEE, 2017: 2755-2763.
[20]	HAN S, MAO H Z, DALLY W J. Deep compression: Compressing deep neural network with pruning, trained quantization and huffman coding[J]. Fiber, 2015, 56(4): 3-7.
[21]	孙文宇, 曹健, 李普, 等. 基于全局信息的卷积神经网络模型剪枝微调优化方法[J]. 北京大学学报(自然科学版), 2021, 57(4): 790-794.
	SUN Wenyu, CAO Jian, LI Pu, et al. Pruning and fine-tuning optimization method of convolutional neural network based on global information[J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 2021, 57(4): 790-794.
[22]	GONG H, LI H, XU K, et al. Object detection based on improved YOLOv3-tiny[C]// 2019 Chinese Automation Congress. Hangzhou, China: IEEE, 2019: 3240-3245.
[23]	GE Z, LIU S T, WANG F, et al. YOLOX: Exceeding YOLO series in 2021[J/OL].(2021-07-18)[2021-12-06]. https://arxiv.org/abs/2107.08430.