基于元学习的小样本指针式仪表检测方法

doi:10.11996/JG.j.2095-302X.2025010081

图学学报 ›› 2025, Vol. 46 ›› Issue (1): 81-93.DOI: 10.11996/JG.j.2095-302X.2025010081

• 图像处理与计算机视觉 • 上一篇下一篇

基于元学习的小样本指针式仪表检测方法

孙前来(), 林绍杭, 刘东峰, 宋晓阳, 刘佳耀, 刘瑞珍

太原科技大学电子信息工程学院，山西太原 030024

收稿日期:2024-06-23 接受日期:2024-10-28 出版日期:2025-02-28 发布日期:2025-02-14
第一作者:孙前来(1976-)，男，副教授，博士。主要研究方向为智能控制理论及应用、机器视觉。E-mail：2000025@tyust.edu.cn
基金资助:
山西省重点研发计划项目(202102020101005);山西省高等学校科技创新项目(2023L185);来晋工作优秀博士奖励资金(20222088);太原科技大学研究生教育创新项目(SY2023016)

Few-shot pointer meters detection method based on meta-learning

SUN Qianlai(), LIN Shaohang, LIU Dongfeng, SONG Xiaoyang, LIU Jiayao, LIU Ruizhen

School of Electronic Information Engineering, Taiyuan University of Science and Technology, Taiyuan Shanxi 030024, China

Received:2024-06-23 Accepted:2024-10-28 Published:2025-02-28 Online:2025-02-14
First author：SUN Qianlai (1976-), associate professor, Ph.D. His main research interests cover include intelligent control theory and applications, machine vision. E-mail：2000025@tyust.edu.cn
Supported by:
Key Research and Developmen Program Projects in Shanxi Province(202102020101005);Shanxi Province Higher Education Science and Technology Innovation Project(2023L185);Reward Funds for Outstanding Doctoral Students Working in Shanxi(20222088);Postgraduate Education and Innovation Project in Taiyuan University of Science and Technology(SY2023016)

摘要/Abstract

摘要：

仪表定位精度是保证指针式仪表示数准确识别的前提。复杂工业场景下仪表样本难以采集，小样本情况下，现有指针式仪表定位检测方法存在检测精度低、实时性差的问题。为此，提出了基于元学习的小样本指针式仪表检测方法Sparse-Meta-DETR。并以Meta-DETR为目标检测基线模型，采用元学习训练策略，在元训练阶段构建多个小样本任务训练Sparse-Meta-DETR模型，增强特征相关聚合模块对特征空间中支持集和查询集类别的余弦距离的度量能力，使模型能够在元测试阶段小样本任务中识别图像包含的类别，快速适应新类小样本任务，检测复杂工业场景图像中包含的指针式仪表；引入轻量级主干网络Efficientnet b1作为特征提取器，减少模型的计算复杂度和参数量；设计评分网络对查询特征稀疏采样，构建稀疏化遮罩选取前景特征，引导Transformer编/解码器对前景特征进行处理，进一步减少计算量并提高检测精度。使用Sparse-Meta-DETR模型，20-shot时指针式仪表定位检测精度指标AP⁵⁰和AP⁷⁵分别达到了94.2%和87.5%，10-shot时的AP⁵⁰达到了91.1%；相较于最初的基线模型，改进模型的时间复杂度下降了74.5%。实验结果表明，Sparse-Meta-DETR不仅能够保证样本匮乏时仪表定位的精度，还可以有效地提高仪表定位的实时性，其整体性能优于Meta-RCNN等小样本深度学习算法。

关键词: 指针式仪表, 元学习, 小样本, 目标检测, 稀疏采样

Abstract:

The accuracy of meters location is a critical factor in ensuring the accuracy of meters recognition. However, it is challenging to collect instrument data in complex industrial scenarios, existing pointer instrument detection methods exhibit low detection accuracy and poor real-time performance in few-shot situations. For this reason, the Sparse-Meta-DETR method was proposed for few-shot pointer meter detection based on meta-learning. Inspired by the object detection model Meta-DETR, this method adopted the meta-learning strategy. During the meta-training stage, few-shot tasks were created to train the Sparse-Meta-DETR model, enhancing metrics ability of the correlational aggregation module for support set and query set classes in the feature space. This enabled the model to recognize classes present in images during the few-shot training stage with few-shot tasks, quickly adapt to few-shot tasks with novel classes, detect pointer meters in complex industrial scenarios. A lightweight backbone network, Efficientnet b1, was introduced as the feature extractor to reduce the computational complexity and parameter of the model, thereby improving the detection speed. Simultaneously, a scoring network was designed as a token sparsification sampler, creating a sparsification mask to select foreground features from query features. This guided the Transformer encoders and decoders to focus on foreground features, thereby reducing computational complexity of few-shot training stage and improving detection accuracy. The Sparse-Meta-DETR model achieved an AP⁵⁰of 94.2% and an AP⁷⁵ of 87.5% in 20-shot task, and an AP⁵⁰ of 91.1% in 10-shot tasks. Compared to the baseline model, the improved model reduced time complexity by 74.5%. Experimental results demonstrated that the Sparse-Meta-DETR can not only effectively ensure the accuracy of pointer meter positioning detection but also improve the real-time performance in the case of few-shot. Its overall performance surpassed other few-shot deep-learning algorithms such as Meta RCNN.

Key words: pointer meter, meta-learning, few-shot, object detection, sparsification

中图分类号:

孙前来, 林绍杭, 刘东峰, 宋晓阳, 刘佳耀, 刘瑞珍. 基于元学习的小样本指针式仪表检测方法[J]. 图学学报, 2025, 46(1): 81-93.

SUN Qianlai, LIN Shaohang, LIU Dongfeng, SONG Xiaoyang, LIU Jiayao, LIU Ruizhen. Few-shot pointer meters detection method based on meta-learning[J]. Journal of Graphics, 2025, 46(1): 81-93.

图/表 12

参考文献 37

[1]	郝琨, 韩冰, 李志圣, 等. 基于投影阈值分割和数字序列校正的高噪声数字仪表图像识别方法[J]. 电子科技大学学报, 2023, 52(5): 728-738.
	HAO K, HAN B, LI Z S, et al. High noise digital instrument image recognition method based on projection threshold segmentation and number sequence correction[J]. Journal of University of Electronic Science and Technology of China, 2023, 52(5): 728-738 (in Chinese).
[2]	毛爱坤, 刘昕明, 陈文壮, 等. 改进YOLOv5算法的变电站仪表目标检测方法[J]. 图学学报, 2023, 44(3): 448-455. DOI
	MAO A K, LIU X M, CHEN W Z, et al. Improved substation instrument target detection method for YOLOv5 algorithm[J]. Journal of Graphics, 2023, 44(3): 448-455 (in Chinese).
[3]	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).
[4]	赵振兵, 马迪雅, 石颖, 等. 基于改进YOLOX的变电站仪表外观缺陷检测算法[J]. 图学学报, 2023, 44(5): 937-946. DOI
	ZHAO Z B, MA D Y, SHI Y, et al. Appearance defect detection algorithm of substation instrument based on improved YOLOX[J]. Journal of Graphics, 2023, 44(5): 937-946 (in Chinese).
[5]	任志玲, 曹正言, 任立然. 基于SCC-YOLO的指针式仪表轻量化检测方法[J]. 仪表技术与传感器, 2024(9): 39-47, 52.
	REN Z L, CAO Z Y, REN L R. Lightweight detection method for pointer meters based on SCC-YOLO[J]. Instrument Technique and Sensor, 2024(9): 39-47, 52 (in Chinese).
[6]	刘萌, 王波, 罗鹏, 等. 适应于户外场景下低成像质量的指针式表计读数智能识别方法[EB/OL]. (2024-07-24) [2024-08-22]. https://doi.org/10.13336/j.1003-6520.hve.20231408.
	LIU M, WANG B, LUO P, et al. Intelligent recognition method of pointer meter reading suitable for low imaging quality in outdoor scene[EB/OL]. (2024-07-24) [2024-08-12]. https://doi.org/10.13336/j.1003-6520.hve.20231408 (in Chinese).
[7]	赵丽娜, 赵丽娟, 宁子达, 等. 一种SF₆压力指针式仪表的定位识别方法[EB/OL]. (2022-11-03) [2024-04-19]. https://kns.cnki.net/kcms/detail/23.1202.TH.20221103.1102.002.html.
	ZHAO L N, ZHAO L J, NING Z D, et al. A positioning recognition method for SF₆ pressure pointer meter[EB/OL]. (2022-11-03) [2024-04-19]. http://kns.cnki.net/kcms/detail/23.1202.TH.20221103.1102.002.html (in Chinese).
[8]	徐小强, 陈旭, 冒燕. 基于戴明回归和ROI细化的指针式仪表读数技术[J]. 江苏大学学报(自然科学版), 2023, 44(3): 324-329.
	XU X Q, CHEN X, MAO Y. Reading technology of pointer instrument based on Deming regression and ROI thinning[J]. Journal of Jiangsu University (Natural Science Edition), 2023, 44(3): 324-329 (in Chinese).
[9]	王欣然, 张斌, 湛敏, 等. 基于改进的Faster RCNN的仪表自动识别方法[J]. 机电工程, 2024, 41(3): 532-539.
	WANG X R, ZHANG B, ZHAN M, et al. Automatic instrument identification based on improved Faster RCNN[J]. Journal of Mechanical & Electrical Engineering, 2024, 41(3): 532-539 (in Chinese).
[10]	刘海龙, 马波, 张奇志. 钻井现场指针式仪表读数自动识别方法研究[J]. 控制工程, 2023, 30(9): 1665-1672.
	LIU H L, MA B, ZHANG Q Z. Study on automatic identification method of pointer instrument reading in drilling field[J]. Control Engineering of China, 2023, 30(9): 1665-1672 (in Chinese).
[11]	黄思远, 樊绍胜, 王子扬. 基于CenterNet和DeepLabv3+的变电站指针式仪表读数识别方法[J]. 电力学报, 2022, 37(3): 232-243.
	HUANG S Y, FAN S S, WANG Z Y. Substation pointer meter reading recognition method based on CenterNet and DeepLabv3+[J]. Journal of Electric Power, 2022, 37(3): 232-243 (in Chinese).
[12]	ZHANG H Y, RAO Y B, SHAO J, et al. Reading various types of pointer meters under extreme motion blur[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 5019815.
[13]	KAUL P, XIE W D, ZISSERMAN A, et al. Label, verify, correct: a simple few shot object detection method[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2022: 14217-14227.
[14]	GUIRGUIS K, HENDAWY A, ESKANDAR G, et al. CFA: constraint-based finetuning approach for generalized few-shot object detection[C]// 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops. New York: IEEE Press, 2022: 4038-4048.
[15]	LI B H, YANG B Y, LIU C, et al. Beyond max-margin: class margin equilibrium for few-shot object detection[C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York: IEEE Press, 2021: 7359-7368.
[16]	YAN X P, CHEN Z L, XU A N, et al. Meta R-CNN: towards general solver for instance-level low-shot learning[C]// 2019 IEEE/CVF International Conference on Computer Vision. New York: IEEE Press, 2019: 9576-9585.
[17]	HAN G X, HUANG S Y, MA J W, et al. Meta faster R-CNN: towards accurate few-shot object detection with attentive feature alignment[C]// The 36th AAAI Conference on Artificial Intelligence. Palo Alto: AAAI Press, 2022: 780-789.
[18]	ZHANG G J, LUO Z P, CUI K W, et al. Meta-DETR: image-level few-shot detection with inter-class correlation exploitation[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(11): 12832-12843.
[19]	翟永杰, 赵振远, 王乾铭, 等. 基于人工-真实样本度量学习的指针式仪表检测方法[J]. 电测与仪表, 2022, 59(10): 174-183.
	ZHAI Y J, ZHAO Z Y, WANG Q M, et al. Pointer meter detection method based on artificial-real sample metric learning[J]. Electrical Measurement & Instrumentation, 2022, 59(10): 174-183 (in Chinese).
[20]	史燕燕, 史殿习, 乔子腾, 等. 小样本目标检测研究综述[J]. 计算机学报, 2023, 46(8): 1753-1780.
	SHI Y Y, SHI D X, QIAO Z T, et al. A survey on recent advances in few-shot object detection[J]. Chinese Journal of Computers, 2023, 46(8): 1753-1780 (in Chinese).
[21]	O'SHEA K, NASH R. An introduction to convolutional neural networks[EB/OL]. (2015-11-02) [2024-04-06]. https://arxiv.org/abs/1511.08458.
[22]	SCABINI L F S, BRUNO O M. Structure and performance of fully connected neural networks: emerging complex network properties[EB/OL]. (2021-07-29) [2024-04-06]. https://arxiv.org/abs/2107.14062.
[23]	VINYALS O, BLUNDELL C, LILLICRAP T, et al. Matching networks for one shot learning[C]// The 30th International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2016: 3637-3645.
[24]	LI J, CHENG J H, SHI J Y, et al. Brief introduction of back propagation (BP) neural network algorithm and its improvement[C]// Advances in Computer Science and Information Engineering. Cham: Springer, 2012: 553-558.
[25]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C]// The 31st International Conference on Neural Information Processing Systems. Red Hook: Curran Associates Inc., 2017: 6000-6010.
[26]	张阳婷, 黄德启, 王东伟, 等. 基于深度学习的目标检测算法研究与应用综述[J]. 计算机工程与应用, 2023, 59(18): 1-13. DOI
	ZHANG Y T, HUANG D Q, WANG D W, et al. Review on research and application of deep learning-based target detection algorithms[J]. Computer Engineering and Applications, 2023, 59(18): 1-13 (in Chinese). DOI
[27]	TAN M X, LE Q. EfficientNet: rethinking model scaling for convolutional neural networks[EB/OL]. (2020-09-11) [2024-04-19]. http://proceedings.mlr.press/v97/tan19a.html.
[28]	BA J L, KIROS J R, HINTON G E. Layer normalization[EB/OL]. (2016-07-21) [2024-04-19]. https://arxiv.org/abs/1607.06450.
[29]	KUNC V, KLÉMA J. Three decades of activations: a comprehensive survey of 400 activation functions for neural networks[EB/OL]. (2024-02-14) [2024-04-19]. https://arxiv.org/abs/2402.09092.
[30]	CARION N, MASSA F, SYNNAEVE G, et al. End-to-end object detection with transformers[C]// The 16th European Conference on Computer Vision. Cham: Springer, 2020: 213-229.
[31]	ROH B, SHIN J W, SHIN W, et al. Sparse DETR: efficient end-to-end object detection with learnable sparsity[EB/OL]. (2022-03-04) [2024-04-19]. https://openreview.net/forum?id=RRGVCN8kjim.
[32]	MA Y D, LIU Q, QIAN Z B. Automated image segmentation using improved PCNN model based on cross-entropy[C]// 2004 International Symposium on Intelligent Multimedia, Video and Speech Processing. New York: IEEE Press, 2004: 743-746.
[33]	CHEN W Y, LIU Y C, KIRA Z, et al. A closer look at few-shot classification[EB/OL]. (2020-01-12) [2024-04-19]. https://openreview.net/forum?id=HkxLXnAcFQ.
[34]	LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]// 2017 IEEE International Conference on Computer Vision. New York: IEEE Press, 2017: 2999-3007.
[35]	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. New York: IEEE Press, 2019: 658-666.
[36]	EVERINGHAM M, VAN GOOL L, WILLIAMS C K I, et al. The PASCAL visual object classes (VOC) challenge[J]. International Journal of Computer Vision, 2010, 88(2): 303-338.
[37]	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.

方法	代表模型	特点
基于元学习的方法	Meta-RCNN^[16] Meta-DETR^[18]	检测精度高，适用于样本数量极少的任务
基于迁移学习的方法	CFA-DeFRCN^[14]	易于实现，但不适用于样本数量极少的任务
基于数据增强的方法	LVC^[13]	易于实现，但实时性较差，定位精度较低
基于度量学习的方法	CME^[15] ARSML^[19]	适用于样本数量极少的任务，但定位精度低

方法	代表模型	特点
基于元学习的方法	Meta-RCNN^[16] Meta-DETR^[18]	检测精度高，适用于样本数量极少的任务
基于迁移学习的方法	CFA-DeFRCN^[14]	易于实现，但不适用于样本数量极少的任务
基于数据增强的方法	LVC^[13]	易于实现，但实时性较差，定位精度较低
基于度量学习的方法	CME^[15] ARSML^[19]	适用于样本数量极少的任务，但定位精度低

Stage	卷积层	输入尺寸	通道数	层数
1	Conv 3×3	224×224	32	1
2	MBConv1, k3×3	112×112	16	1
3	MBConv6, k3×3	112×112	24	2
4	MBConv6, k3×3	56×56	40	2
5	MBConv6, k3×3	28×28	80	3
6	MBConv6, k3×3	14×14	112	3
7	MBConv6, k3×3	14×14	192	4
8	MBConv6, k3×3	7×7	320	1
9	Conv 1×1 & Pooling & FC	7×7	1280	1

Stage	卷积层	输入尺寸	通道数	层数
1	Conv 3×3	224×224	32	1
2	MBConv1, k3×3	112×112	16	1
3	MBConv6, k3×3	112×112	24	2
4	MBConv6, k3×3	56×56	40	2
5	MBConv6, k3×3	28×28	80	3
6	MBConv6, k3×3	14×14	112	3
7	MBConv6, k3×3	14×14	192	4
8	MBConv6, k3×3	7×7	320	1
9	Conv 1×1 & Pooling & FC	7×7	1280	1

操作名称	输入尺寸	输出尺寸	层数
Layer normalization	l×d	l×d	1
Linear layer & GELU	l×d	l×d	1
Linear layer & GELU	l×d	l×d/2	1
Linear layer & GELU	l×d/2	l×d/4	1
Linear layer	l×d/4	l×1	1

基于元学习的小样本指针式仪表检测方法

Few-shot pointer meters detection method based on meta-learning

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 12

参考文献 37

相关文章 15

编辑推荐

Metrics

本文评价

名称	参数
CPU	Intel Xeon Gold 6133
内存	32 G
GPU	NVIDIA GeForce RTX A5000 24 G
操作系统	Ubuntu 18.04 LTS
CUDA	11.3
深度学习框架	Pytorch-1.12.0
编程语言	Python3.8

Model	shot	AP⁵⁰	AP⁷⁵	AP^50_95	Macs/G	Param/M
ARSML	1	0.762	0.224	0.409	-	-
	10	0.892	0.617	0.550
	20	0.934	0.778	0.619
Meta-RCNN	1	0.323	-	-	35.62	45.93
	10	0.833	-	-
	20	0.908	-	-
CFA-DeFRCN	1	0.809	0.114	0.376	30.47	42.06
	10	0.882	0.546	0.505
	20	0.930	0.762	0.646
LVC	1	0.672	0.134	0.352	31.54	43.17
	10	0.864	0.519	0.496
	20	0.923	0.749	0.630
baseline	1	0.854	0.560	0.523	12.74	28.10
	10	0.949	0.925	0.707
	20	0.978	0.920	0.722
proposed	1	0.841	0.156	0.369	3.24	9.27
	10	0.912	0.802	0.629
	20	0.942	0.875	0.696

Model	轻量化主干网络	shot	AP⁵⁰	AP⁷⁵	AP^50_95	Macs/G	Param/M
Meta-DETR-R50(baseline)		1	0.854	0.560	0.523	12.74	28.10
		10	0.949	0.925	0.707
		20	0.978	0.920	0.722
Meta-DETR-Mobv2	√	1	0.216	0.166	0.135	2.69	5.96
		10	0.833	0.652	0.523
		20	0.875	0.700	0.550
Meta-DETR-Effb0	√	1	0.655	0.351	0.359	2.95	7.01
		10	0.867	0.565	0.520
		20	0.901	0.791	0.627
Meta-DETR-Effb1	√	1	0.781	0.346	0.411	3.47	8.56
		10	0.865	0.723	0.587
		20	0.934	0.866	0.705
Sparse-Meta-DETR-Effb1(proposed)	√	1	0.841	0.156	0.369	3.24	9.27
		10	0.912	0.802	0.629
		20	0.942	0.875	0.696

[1]	程旭东, 史彩娟, 高炜翔, 王森, 段昌钰, 闫晓东. 面向域自适应目标检测的一致无偏教师模型[J]. 图学学报, 2025, 46(1): 114-125.
[2]	崔克彬, 耿佳昌. 基于EE-YOLOv8s的多场景火灾迹象检测算法[J]. 图学学报, 2025, 46(1): 13-27.
[3]	苑朝, 赵明雪, 张丰羿, 冯晓勇, 李冰, 陈瑞. 基于点云特征增强的复杂室内场景3D目标检测[J]. 图学学报, 2025, 46(1): 59-69.
[4]	王杨, 马唱, 胡明, 孙涛, 饶元, 袁振羽. 基于多尺度特征融合的轻量型野外蝙蝠检测[J]. 图学学报, 2025, 46(1): 70-80.
[5]	李琼, 考月英, 张莹, 徐沛. 面向无人机航拍图像的目标检测研究综述[J]. 图学学报, 2024, 45(6): 1145-1164.
[6]	李珍峰, 符世琛, 徐乐, 孟博, 张昕, 秦建军. 基于MBI-YOLOv8的煤矸石目标检测算法研究[J]. 图学学报, 2024, 45(6): 1301-1312.
[7]	李盛涛, 侯立群, 董亚松. 基于R-YOLOv7和MIMO-CTFNet的指针式仪表自动读数方法[J]. 图学学报, 2024, 45(6): 1313-1327.
[8]	闫建红, 冉同霄. 基于YOLOv8的轻量化无人机图像目标检测算法[J]. 图学学报, 2024, 45(6): 1328-1337.
[9]	姜晓恒, 段金忠, 卢洋, 崔丽莎, 徐明亮. 融合先验知识推理的表面缺陷检测[J]. 图学学报, 2024, 45(5): 957-967.
[10]	章东平, 魏杨悦, 何数技, 徐云超, 胡海苗, 黄文君. 特征融合与层间传递：一种基于Anchor DETR改进的目标检测方法[J]. 图学学报, 2024, 45(5): 968-978.
[11]	李建华, 韩宇, 石开铭, 张可嘉, 郭红领, 方东平, 曹佳明. 施工现场小目标工人检测方法[J]. 图学学报, 2024, 45(5): 1040-1049.
[12]	孙己龙, 刘勇, 周黎伟, 路鑫, 侯小龙, 王亚琼, 王志丰. 基于DCNv2和Transformer Decoder的隧道衬砌裂缝高效检测模型研究[J]. 图学学报, 2024, 45(5): 1050-1061.
[13]	李大湘, 吉展, 刘颖, 唐垚. 改进YOLOv7遥感图像目标检测算法[J]. 图学学报, 2024, 45(4): 650-658.
[14]	胡欣, 常娅姝, 秦皓, 肖剑, 程鸿亮. 基于改进YOLOv8和GMM图像点集匹配的双目测距方法[J]. 图学学报, 2024, 45(4): 714-725.
[15]	牛为华, 郭迅. 基于改进YOLOv8的船舰遥感图像旋转目标检测算法[J]. 图学学报, 2024, 45(4): 726-735.