用于半监督火灾检测的分布自适应和动态课程伪标签框架

doi:10.11772/j.issn.1001-9081.2024101452

《计算机应用》唯一官方网站 ›› 2025, Vol. 45 ›› Issue (10): 3131-3137.DOI: 10.11772/j.issn.1001-9081.2024101452

• 人工智能 • 上一篇

用于半监督火灾检测的分布自适应和动态课程伪标签框架

王磊, 胡节(), 彭博

西南交通大学计算机与人工智能学院，成都 611756

收稿日期:2024-10-14 修回日期:2024-12-16 接受日期:2024-12-20 发布日期:2024-12-24 出版日期:2025-10-10
通讯作者: 胡节
作者简介:王磊（2000—），男，四川泸州人，硕士，CCF会员，主要研究方向：图像目标检测
胡节（1978—），女，四川成都人，副教授，博士，CCF会员，主要研究方向：时间序列预测、图像目标检测、知识图谱 Email:jiehu@swjtu.edu.cn
彭博（1980—），女，四川成都人，教授，博士，主要研究方向：图像分割、图像目标识别。
基金资助:
四川省重点研发计划项目(2023YFG0354)

Distribution adaptation and dynamic curriculum pseudo-label framework for semi-supervised fire detection

Lei WANG, Jie HU(), Bo PENG

School of Computing and Artificial Intelligence，Southwest Jiaotong University，Chengdu Sichuan 611756，China

Received:2024-10-14 Revised:2024-12-16 Accepted:2024-12-20 Online:2024-12-24 Published:2025-10-10
Contact: Jie HU
About author:WANG Lei， born in 2000， M. S. His research interests include image object detection.
HU Jie， born in 1979， Ph. D.， associate professor. Her research interests include time series prediction， image object detection， knowledge graph.
PENG Bo， born in 1980， Ph. D.， professor. Her research interests include image segmentation， image object recognition.
Supported by:
Sichuan Key Research and Development Program(2023YFG0354)

摘要/Abstract

摘要：

针对火灾图像标签过少、背景复杂多样导致的半监督目标检测困难的问题，提出一种用于半监督火灾检测的分布自适应和动态课程伪标签框架（DADCPL-SFD）。该框架主要由师生互学（ML）框架、软标签（SL）、分布自适应和动态课程伪标签这4个部分组成。首先，采用师生互学框架的半监督学习范式替换YOLOv5-l的全监督学习范式，以应对数据标签少的场景；其次，采用软标签以获取更多有效的伪标签正例，优化半监督学习过程；再次，引入分布自适应损失，减小源域和目标域的数据分布差异，使模型在不同域上表现一致；最后，设计一种基于课程思想的动态课程伪标签策略，根据伪标签生成的情况在不同训练时期调整阈值，以筛选更合理的伪标签。在火焰和烟雾数据集（DFS）多个监督比例上（1%、2%、5%和10%）的实验结果表明，相较于全监督学习，所提框架的平均精度均值（mAP）平均提升了5.32个百分点，在交并比（IoU）阈值为0.5下的平均精度（AP）平均提升了11.87个百分点，充分验证了DADCPL-SFD的高效性和准确性。

关键词: 火灾检测, 半监督目标检测, 师生互学, 软标签, 分布自适应, 动态课程伪标签

Abstract:

To address the challenges in semi-supervised object detection due to the lack of fire image labels and the complexity and diversity of background， a Distribution Adaptation and Dynamic Curriculum Pseudo-Label framework for Semi-supervised Fire Detection （DADCPL-SFD） was proposed， which consisted of four parts： teacher-student Mutual Learning （ML） framework， Soft Label （SL）， distribution adaptation and dynamic curriculum pseudo-label. Firstly， the semi-supervised learning paradigm of teacher-student mutual learning framework was adopted to replace the fully supervised learning paradigm of YOLOv5-l for the scenario with few data labels. Secondly， soft labels were used to obtain more effective pseudo-label positive examples and optimize the semi-supervised learning process. Thirdly， the distribution adaptation loss was introduced to reduce data distribution difference between the source domain and the target domain， thereby ensuring consistent model performance across different domains. Finally， a dynamic curriculum pseudo-label strategy， inspired by the concept of curriculum learning， was designed to adjust the threshold according to the pseudo-label generation condition dynamically during different training periods， so as to filter more reasonable pseudo-labels. Experimental results on the Dataset for Fire and Smoke detection （DFS） at various supervision ratios （1%， 2%， 5%， and 10%） show that compared to the supervised learning， the proposed framework has the mean Average Precision （mAP） improved by an average of 5.32 percentage points， and the Average Precision （AP） at an Intersection over Union （IoU） threshold of 0.5 improved by an average of 11.87 percentage points， which fully demonstrates the efficiency and accuracy of DADCPL-SFD.

Key words: fire detection, semi-supervised object detection, teacher-student Mutual Learning (ML), Soft Label (SL), distribution adaptation, dynamic curriculum pseudo-label

中图分类号:

TP391.4

王磊, 胡节, 彭博. 用于半监督火灾检测的分布自适应和动态课程伪标签框架[J]. 计算机应用, 2025, 45(10): 3131-3137.

Lei WANG, Jie HU, Bo PENG. Distribution adaptation and dynamic curriculum pseudo-label framework for semi-supervised fire detection[J]. Journal of Computer Applications, 2025, 45(10): 3131-3137.

图/表 8

参考文献 21

[1]	TALAAT F M， ZAINELDIN H. An improved fire detection approach based on YOLO-v8 for smart cities［J］. Neural Computing and Applications， 2023， 35（28）： 20939-20954.
[2]	LI C， LI G， SONG Y， et al. Fast forest fire detection and segmentation application for UAV-assisted mobile edge computing system［J］. IEEE Internet of Things Journal， 2024， 11（16）： 26690-26699.
[3]	DIWAN T， ANIRUDH G， TEMBHURNE J V. Object detection using YOLO： challenges， architectural successors， datasets and applications［J］. Multimedia Tools and Applications， 2023， 82（6）： 9243-9275.
[4]	AMIT Y， FELZENSZWALB P， GIRSHICK R. Object detection［M］// IKEUCHI K. Computer vision： a reference guide， 2nd ed. Cham： Springer， 2021： 875-883.
[5]	ROH J， KIM Y， KONG M. Fire image classification based on convolutional neural network for smart fire detection［J］. International Journal of Fire Science and Engineering， 2022， 36（3）： 51-61.
[6]	MAJID S， ALENEZI F， MASOOD S， et al. Attention based CNN model for fire detection and localization in real-world images［J］. Expert Systems with Applications， 2022， 189： No.116114.
[7]	陈佳慧，王晓虹. 改进YOLOv5的无人机航拍图像密集小目标检测算法［J］. 计算机工程与应用， 2024， 60（3）：100-108.
	CHEN J H， WANG X H. Dense small object detection algorithm based on improved YOLOv5 in UAV aerial images［J］. Computer Engineering and Applications， 2024， 60（3）： 100-108.
[8]	李军，刘念，张世义. 基于YOLOv5的桥梁裂纹检测方法研究［J］. 计算机科学， 2024， 51（11A）： No.231200063.
	LI J， LIU N， ZHANG S Y. Study on detection method of bridge crack based on YOLOv5［J］. Computer Science， 2024， 51（11A）： No.231200063.
[9]	YAR H， KHAN Z A， ULLAH F U M， et al. A modified YOLOv5 architecture for efficient fire detection in smart cities［J］. Expert Systems with Applications， 2023， 231： No.120465.
[10]	TARVAINEN A， VALPOLA H. Mean teachers are better role models： weight-averaged consistency targets improve semi-supervised deep learning results［C］// Proceedings of the 31st International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2017： 1195-1204.
[11]	LIU Y C， MA C Y， KIRA Z. Unbiased Teacher for semi-supervised object detection［EB/OL］. ［2024-12-05］. .
[12]	LIU Y C， MA C Y， KIRA Z. Unbiased Teacher v2： semi-supervised object detection for anchor-free and anchor-based detectors［C］// Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2022： 9809-9818.
[13]	LIN Q， LI Z， ZENG K， et al. FireMatch： a semi-supervised video fire detection network based on consistency and distribution alignment［J］. Expert Systems with Applications， 2024， 248： No.123409.
[14]	SUN G， WEN Y， LI Y. Instance segmentation using semi-supervised learning for fire recognition［J］. Heliyon， 2022， 8（12）： No.e12375.
[15]	WANG X， ZHANG R， KONG T， et al. SOLOv2： dynamic and fast instance segmentation［C］// Proceedings of the 34th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2020： 17721-17732.
[16]	WANG C， GRAU A， GUERRA E， et al. Semi-supervised wildfire smoke detection based on smoke-aware consistency［J］. Frontiers in Plant Science， 2022， 13： No.980425.
[17]	ZHANG B， WANG Y， HOU W， et al. FlexMatch： boosting semi-supervised learning with curriculum pseudo labeling［C］// Proceedings of the 35th International Conference on Neural Information Processing Systems. Red Hook： Curran Associates Inc.， 2021： 18408-18419.
[18]	WU S， ZHANG X， LIU R， et al. A dataset for fire and smoke object detection［J］. Multimedia Tools and Applications， 2023， 82（5）： 6707-6726.
[19]	CHEN B， CHEN W， YANG S， et al. Label matching semi-supervised object detection［C］// Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway： IEEE， 2022： 14361-14370.
[20]	SOHN K， ZHANG Z， LI C L， et al. A simple semi-supervised learning framework for object detection［EB/OL］. ［2024-12-05］. .
[21]	XU M， ZHANG Z， HU H， et al. End-to-end semi-supervised object detection with soft teacher［C］// Proceedings of the 2021 IEEE/CVF International Conference on Computer Vision. Piscataway： IEEE， 2021： 3040-3049.

监督比例/%	样本数
监督比例/%	有标注集	无标注集	验证集
共计	9 462
1	65	6 559	2 838
2	129	6 495	2 838
5	320	6 304	2 838
10	623	6 001	2 838

监督比例/%	样本数
监督比例/%	有标注集	无标注集	验证集
共计	9 462
1	65	6 559	2 838
2	129	6 495	2 838
5	320	6 304	2 838
10	623	6 001	2 838

监督比例	算法	AP₅₀	mAP
1	Unbiased Teacher^［11］	9.87	2.87
	Unbiased Teacher v2^［12］	12.90	4.88
	LabelMatch^［19］	11.90	4.03
	STAC^［20］	9.90	2.90
	SoftTeacher^［21］	12.80	4.00
	DADCPL-SFD	15.20	5.19
2	Unbiased Teacher^［11］	10.80	3.18
	Unbiased Teacher v2^［12］	12.20	3.77
	LabelMatch^［19］	15.20	5.19
	STAC^［20］	13.90	4.30
	SoftTeacher^［21］	20.60	6.50
	DADCPL-SFD	20.30	7.94
5	Unbiased Teacher^［11］	13.90	4.35
	Unbiased Teacher v2^［12］	13.00	4.20
	LabelMatch^［19］	23.80	9.76
	STAC^［20］	20.50	6.60
	SoftTeacher^［21］	20.40	6.64
	DADCPL-SFD	26.90	10.90
10	Unbiased Teacher^［11］	25.90	10.10
	Unbiased Teacher v2^［12］	26.20	10.20
	LabelMatch^［19］	30.60	12.50
	STAC^［20］	23.70	8.50
	SoftTeacher^［21］	21.40	7.10
	DADCPL-SFD	33.00	12.80

监督比例	算法	AP₅₀	mAP
1	Unbiased Teacher^［11］	9.87	2.87
	Unbiased Teacher v2^［12］	12.90	4.88
	LabelMatch^［19］	11.90	4.03
	STAC^［20］	9.90	2.90
	SoftTeacher^［21］	12.80	4.00
	DADCPL-SFD	15.20	5.19
2	Unbiased Teacher^［11］	10.80	3.18
	Unbiased Teacher v2^［12］	12.20	3.77
	LabelMatch^［19］	15.20	5.19
	STAC^［20］	13.90	4.30
	SoftTeacher^［21］	20.60	6.50
	DADCPL-SFD	20.30	7.94
5	Unbiased Teacher^［11］	13.90	4.35
	Unbiased Teacher v2^［12］	13.00	4.20
	LabelMatch^［19］	23.80	9.76
	STAC^［20］	20.50	6.60
	SoftTeacher^［21］	20.40	6.64
	DADCPL-SFD	26.90	10.90
10	Unbiased Teacher^［11］	25.90	10.10
	Unbiased Teacher v2^［12］	26.20	10.20
	LabelMatch^［19］	30.60	12.50
	STAC^［20］	23.70	8.50
	SoftTeacher^［21］	21.40	7.10
	DADCPL-SFD	33.00	12.80

模块					监督比例为1%		监督比例为2%		监督比例为5%		监督比例为10%
YOLOv5-l	ML	SL	DA	DCPL	AP₅₀/%	mAP/%	AP₅₀/%	mAP/%	AP₅₀ /%	mAP/%	AP₅₀/%	mAP/%
√					2.08	0.50	7.64	2.24	13.40	4.64	24.80	9.16
√	√				9.87	2.87	10.80	3.18	13.90	4.35	25.90	10.10
√	√	√			12.80	4.51	17.20	6.70	25.50	10.40	30.40	12.90
√	√	√	√		14.00	4.98	19.80	7.74	27.10	10.60	31.50	13.30
√	√	√	√	√	15.20	5.19	20.30	7.94	26.90	10.90	33.00	13.80

用于半监督火灾检测的分布自适应和动态课程伪标签框架

Distribution adaptation and dynamic curriculum pseudo-label framework for semi-supervised fire detection

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 8

参考文献 21

相关文章 1

编辑推荐

Metrics