Improved substation instrument target detection method for YOLOv5 algorithm

doi:10.11996/JG.j.2095-302X.2023030448

Abstract

Abstract:

An efficient and accurate edge instrument detection equipment is crucial for building intelligent substations. However, due to the complex environment of substation, it is challenging for mobile side devices to quickly and accurately detect small targets, multi-categories, and highly similar instrument targets. To address this, a power instrument target detection method based on lightweight SS-YOLOv5 network was proposed. The algorithm was built on YOLOv5 and utilized the lightweight network ShuffleNet V2 to improve the model’s network structure. Deep separable convolutions were introduced to extract instrument features, reducing the computational complexity of the model during neck fusion and improving detection speed. Additionally, combined with Swin Transformer, modeling was undertaken through shift window, allowing for global and local information interaction and enhancing feature extraction ability. Finally, the image data set of substation instrument was constructed independently to train, test, and verify the model. The experimental results demonstrated that, compared with the YOLOv5 algorithm, the model parameters were reduced by 91.8% and the target detection speed was increased by 43.8% for the detection of pressure gauge, ammeter, voltmeter, and other instrument images. It provided a feasible technical scheme for deployment to the side, and could accelerate the development of substation informatization and intelligence.

Key words: deep learning, substation instruments, YOLOv5, target detection, lightweight network

CLC Number:

TP391

MAO Ai-kun, LIU Xin-ming, CHEN Wen-zhuang, SONG Shao-lou. Improved substation instrument target detection method for YOLOv5 algorithm[J]. Journal of Graphics, 2023, 44(3): 448-455.

Figures/Tables 14

References 25

[1]	王刘旺, 周自强, 林龙, 等. 人工智能在变电站运维管理中的应用综述[J]. 高电压技术, 2020, 46(1): 1-13.
	WANG L W, ZHOU Z Q, LIN L, et al. Review on artificial intelligence in substation operation and maintenance management[J]. High Voltage Engineering, 2020, 46(1): 1-13. (in Chinese)
[2]	周振宇, 王曌, 廖海君, 等. 电力物联网5G云-边-端协同框架与资源调度方法[J]. 电网技术, 2022, 46(5): 1641-1651.
	ZHOU Z Y, WANG Z, LIAO H J, et al. 5G cloud-edge-end collaboration framework and resource scheduling method in power Internet of Things[J]. Power System Technology, 2022, 46(5): 1641-1651. (in Chinese)
[3]	邢浩强, 杜志岐, 苏波. 变电站指针式仪表检测与识别方法[J]. 仪器仪表学报, 2017, 38(11): 2813-2821.
	XING H Q, DU Z Q, SU B. Detection and recognition method for pointer-type meter in transformer substation[J]. Chinese Journal of Scientific Instrument, 2017, 38(11): 2813-2821. (in Chinese)
[4]	朱斌滨, 樊绍胜. 雨雾环境下的变电站指针式仪表识别方法[J]. 激光与光电子学进展, 2021, 58(24): 221-230.
	ZHU B B, FAN S S. Reading method of substation pointer meter in rain-fog environment[J]. Laser & Optoelectronics Progress, 2021, 58(24): 221-230. (in Chinese)
[5]	黄山, 吴振升, 任志刚, 等. 电力智能巡检机器人研究综述[J]. 电测与仪表, 2020, 57(2): 26-38.
	HUANG S, WU Z S, REN Z G, et al. Review of electric power intelligent inspection robot[J]. Electrical Measurement & Instrumentation, 2020, 57(2): 26-38. (in Chinese)
[6]	吴杰. 变电站巡检机器人的指针式仪表智能读数系统研究[D]. 武汉: 武汉科技大学, 2019.
	WU J. Research on intelligent recognition system of pointer instrument for substation inspection robot[D]. Wuhan: Wuhan University of Science and Technology, 2019. (in Chinese)
[7]	万吉林, 王慧芳, 管敏渊, 等. 基于Faster R-CNN和U-Net的变电站指针式仪表读数自动识别方法[J]. 电网技术, 2020, 44(8): 3097-3105.
	WAN J L, WANG H F, GUAN M Y, et al. An automatic identification for reading of substation pointer-type meters using faster R-CNN and U-net[J]. Power System Technology, 2020, 44(8): 3097-3105. (in Chinese)
[8]	唐标, 李婷, 李博, 等. 基于边缘计算的变电站内智能布控数据分析与展示[J]. 电气应用, 2021, 40(2): 79-84.
	TANG B, LI T, LI B, et al. Analysis and display of intelligent deployment and control data in substations based on edge computing[J]. Electrotechnical Application, 2021, 40(2): 79-84. (in Chinese)
[9]	李斌, 李亚霖, 朱新山, 等. 基于注意力机制与特征平衡的变电站多目标检测[J]. 电网技术, 2022, 46(6): 2122-2132.
	LI B, LI Y L, ZHU X S, et al. Multi-target detection in substation scence based on attention mechanism and feature balance[J]. Power System Technology, 2022, 46(6): 2122-2132. (in Chinese)
[10]	崔昊杨, 张雨阁, 张驯, 等. 基于边端轻量级网络的电力仪表设备检测方法[J]. 电网技术, 2022, 46(3): 1186-1193.
	CUI H Y, ZHANG Y G, ZHANG X, et al. Detection of power instruments equipment based on edge lightweight network[J]. Power System Technology, 2022, 46(3): 1186-1193. (in Chinese)
[11]	范新南, 黄伟盛, 史朋飞, 等. 基于改进YOLOv4的嵌入式变电站仪表检测算法[J]. 图学学报, 2022, 43(3): 396-403.
	FAN X N, HUANG W S, SHI P F, et al. Embedded substation instrument detection algorithm based on improved YOLOv4[J]. Journal of Graphics, 2022, 43(3): 396-403. (in Chinese)
[12]	何永春, 申永伟, 吴涛, 等. 基于注意力机制的多尺度仪表检测[J]. 科学技术与工程, 2021, 21(31): 13430-13438.
	HE Y C, SHEN Y W, WU T, et al. Multi-scale instrument detection based on attention mechanism[J]. Science Technology and Engineering, 2021, 21(31): 13430-13438. (in Chinese)
[13]	马鹏, 樊艳芳. 基于深度迁移学习的小样本智能变电站电力设备部件检测[J]. 电网技术, 2020, 44(3): 1148-1159.
	MA P, FAN Y F. Small sample smart substation power equipment component detection based on deep transfer learning[J]. Power System Technology, 2020, 44(3): 1148-1159. (in Chinese)
[14]	LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot MultiBox detector[M]//Computer Vision - ECCV 2016. Cham: Springer International Publishing, 2016: 21-37.
[15]	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. New York: IEEE Press, 2016: 779-788.
[16]	GIRSHICK R. Fast R-CNN[C]// 2015 IEEE International Conference on Computer Vision. New York: IEEE Press, 2016: 1440-1448.
[17]	JIANG H Z, LEARNED-MILLER E. Face detection with the faster R-CNN[C]// The 12th IEEE International Conference on Automatic Face & Gesture Recognition. New York: IEEE Press, 2017: 650-657.
[18]	IANDOLA F N, HAN S, MOSKEWICZ M W, et al. SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size[EB/OL]. [2022-02-07]. https://arxiv.org/abs/1602.07360.
[19]	HOWARD A G, ZHU M L, CHEN B, et al. MobileNets: efficient convolutional neural networks for mobile vision applications[EB/OL]. [2022-04-04]. https://arxiv.org/abs/1704.04861.
[20]	SANDLER M, HOWARD A, ZHU M L, et al.MobileNetV2: inverted residuals and linear bottlenecks[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2018: 4510-4520.
[21]	HOWARD A, SANDLER M, CHEN B, et al. Searching for MobileNetV3[C]// 2019 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2020: 1314-1324.
[22]	ZHANG X Y, ZHOU X Y, LIN M X, et al. ShuffleNet: an extremely efficient convolutional neural network for mobile devices[C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2018: 6848-6856.
[23]	MA N N, ZHANG X Y, ZHENG H T, et al. ShuffleNet V2: practical guidelines for efficient CNN architecture design[M]// Computer Vision - ECCV 2018. Cham: Springer International Publishing, 2018: 122-138.
[24]	HAN K, WANG Y H, TIAN Q, et al. GhostNet: more features from cheap operations[C]// 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2020: 1577-1586.
[25]	LIU Z, LIN Y T, CAO Y, et al. Swin transformer: hierarchical vision transformer using shifted windows[C]// 2021 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2022: 9992-10002.

层数	网络层	输入尺寸	步长	通道数
0	CBLM	640×640×3	-	16
1	ShuffleLayer	160×160×16	2	64
2	C3SWinTR	80×80×64	-	64
3	ShuffleLayer	80×80×64	2	120
4	C3	40×40×120	-	120
5	ShuffleLayer	40×40×120	2	232
6	C3	20×20×232	-	232
7	DWConv	20×20×232	1	64
8	Upsample	20×20×64	-	-
9	Concat	-	-	-
10	ShuffleLayer	40×40×184	1	184
11	DWConv	40×40×184	1	32
12	Upsample	40×40×32	-	-
13	Concat	-	-	-
14	ShuffleLayer	80×80×96	1	96
15	Conv	80×80×96	2	32
16	Concat	-	-	-
17	ShuffleLayer	40×40×64	1	64
18	Conv	40×40×64	2	64
19	Concat	-	-	-
20	ShuffleLayer	20×20×128	1	128
21	Detect	-	-	-

层数	网络层	输入尺寸	步长	通道数
0	CBLM	640×640×3	-	16
1	ShuffleLayer	160×160×16	2	64
2	C3SWinTR	80×80×64	-	64
3	ShuffleLayer	80×80×64	2	120
4	C3	40×40×120	-	120
5	ShuffleLayer	40×40×120	2	232
6	C3	20×20×232	-	232
7	DWConv	20×20×232	1	64
8	Upsample	20×20×64	-	-
9	Concat	-	-	-
10	ShuffleLayer	40×40×184	1	184
11	DWConv	40×40×184	1	32
12	Upsample	40×40×32	-	-
13	Concat	-	-	-
14	ShuffleLayer	80×80×96	1	96
15	Conv	80×80×96	2	32
16	Concat	-	-	-
17	ShuffleLayer	40×40×64	1	64
18	Conv	40×40×64	2	64
19	Concat	-	-	-
20	ShuffleLayer	20×20×128	1	128
21	Detect	-	-	-

实验平台	配置
操作系统	Ubuntu 18.04
CPU	Intel@Xeon (R) Gold 6230
GPU	NVIDIA Ge Force RTX2080Ti
编程语言	Python 3.7
深度学习框架	Pytorch

实验平台	配置
操作系统	Ubuntu 18.04
CPU	Intel@Xeon (R) Gold 6230
GPU	NVIDIA Ge Force RTX2080Ti
编程语言	Python 3.7
深度学习框架	Pytorch

检测方案	mAP (%)	模型大小(MB)	GFLOGPs	检测速度(ms)	参数量
方案一	91.00	66.30	9.70	9.3	8 679 120
方案二	82.14	22.50	9.60	7.2	6 056 606
方案三	92.30	5.70	7.10	6.5	656 496
方案四	92.80	2.50	3.54	4.7	537 865
方案五	95.60	13.70	16.40	8.9	7 064 698
方案六	91.80	1.27	1.50	3.7	571 094
本文	94.40	1.32	1.70	5.0	575 198