一种无锚框结构的多尺度火灾检测算法

doi:10.19665/j.issn1001-2400.2022.06.014

摘要/Abstract

摘要：

针对复杂背景下多尺度火焰的检测精度低且容易发生误报等问题,提出了一种无锚框结构的新型火灾检测算法。该算法取消了锚框的预设置,采用逐点预测的方式减少网络的超参数,从而有效地减轻了人工先验知识的影响;引入BFP模块优化特征融合,通过对层间信息的整合有效利用了特征的全局信息,增强了多尺度特征的表达能力;设置融合因子控制层间信息传递,在保证特征信息充分融合的同时减弱高层特征的影响,提升了浅层特征对于小目标的学习能力;设计动态采样方式调整训练过程,采用中心采样和置信度原则提升样本点质量,强化了网络对火焰特征的学习效果。在自建数据集上该算法的检测精度达到了96.9%,在公开数据集上也有较好的检测效果。实验结果表明,该算法检测精度高,抗干扰能力强,对于复杂背景下的多尺度火焰具有较好的检测效果,能够较好抑制误报情况的发生,满足实际火灾检测任务的需要。

关键词: 火灾检测, 无锚框网络, 多尺度特征融合, 动态采样

Abstract:

In view of the low detection accuracy of multi-scale flames and false alarms in complex backgrounds,a new fire detection algorithm with an Anchor Free structure is proposed.The algorithm cancels the anchor and adopts a point-by-point prediction method to reduce the hyperparameters of the network,thus effectively reducing the influence of artificial prior knowledge.The BFP module is introduced to optimize feature fusion,and the integration of inter-layer information effectively utilizes the global information on features and enhances the expression ability of multi-scale features.The fusion factor is set to control the information transfer between layers,which ensures the fusion of feature information while reducing the influence of high-level features,and improves the learning ability of shallow features for small targets.A dynamic sampling method is designed to adjust the training process and strengthen the network’s learning effect on flame characteristics by adopting the principle of central sampling and confidence to improve the quality of the sample points.The algorithm not only achieves 96.9% accuracy on the self-built dataset,but also has a good performance on the public fire dataset.Experimental results show that the proposed algorithm has a high detection accuracy and a strong anti-interference ability.The algorithm has a good detection effect for multi-scale flames in complex backgrounds,can better suppress the occurrence of false alarms,and meets the needs of actual fire detection tasks.

Key words: fire detection, anchor free network, multi-scale feature fusion, dynamic sampling

中图分类号:

TP391.41

秦瑞,张为. 一种无锚框结构的多尺度火灾检测算法[J]. 西安电子科技大学学报, 2022, 49(6): 111-119.

QIN Rui,ZHANG Wei. Multi-scale fire detection algorithm with an anchor free structure[J]. Journal of Xidian University, 2022, 49(6): 111-119.

图/表 15

图1

图2

图3

图4

图5

图6

表1

表2

图7

表3

表4

图8

图9

表5

图10

参考文献 20

[1]	FOGGIA P, SAGGESE A, VENTO M. Real-Time Fire Detection for Video-Surveillance Applications Using a Combination of Experts Based on Color,Shape,and Motion[J]. IEEE Transactions on Circuits & Systems for Video Technology, 2015, 25(9):1545-1556.
[2]	HASHEMZADEH M, ZADEMEHDI A. Fire Detection for Video Surveillance Applications Using ICA K-Medoids-Based Color Model and Efficient Spatio-Temporal Visual Features[J]. Expert Systems with Applications, 2019, 130:60-78. doi: 10.1016/j.eswa.2019.04.019
[3]	何爱龙, 陈美娟. 基于多特征融合的视频火灾检测方法[J]. 软件导刊, 2020, 19(7):198-203.
	HE Ailong, CHEN Meijuan. Video Fire Detection Method Based on Multi-feature Fusion[J]. Software Guide, 2020, 19(7):198-203.
[4]	MUHAMMAD K, AHMAD J, LV Z, et al. Efficient Deep CNN-Based Fire Detection and Localization in Video Surveillance Applications[J]. IEEE Transactions on Systems,Man,and Cybernetics:Systems, 2018, 49(7):1419-1434. doi: 10.1109/TSMC.2018.2830099
[5]	张为, 魏晶晶. 嵌入DenseNet结构和空洞卷积模块的改进YOLO v3火灾检测算法[J]. 天津大学学报:自然科学与工程技术版, 2020, 53(9):976-983.
	ZHANG Wei, WEI Jingjing. Improved YOLOv3 Fire Detection Algorithm Embedded in DenseNet Structure and Dilated Convolution Module[J]. Journal of Tianjin University:Science and Technology, 2020, 53(9):976-983.
[6]	宁阳, 杜建超, 韩硕, 等. 改进 DeeplabV3+ 的火焰分割与火情分析方法[J]. 西安电子科技大学学报, 2021, 48(5):38-46.
	NING Yang, DU Jianchao, HAN Shuo, et al. Fire Segmentation Based on the Improved DeeplabV3+ and the Analytical Method for Fire Development[J]. Journal of Xidian University, 2021, 48(5):38-46.
[7]	JEON M, CHOI H S, LEE J, et al. Multi-Scale Prediction for Fire Detection Using Convolutional Neural Network[J]. Fire Technology, 2021, 57(5):2533-2551. doi: 10.1007/s10694-021-01132-y
[8]	CHAO X C, SHANG W, ZHANG F. Information-Guided Flame Detection Based on Faster R-CNN[J]. IEEE Access, 2020, 8:58923-58932. doi: 10.1109/ACCESS.2020.2982994
[9]	TIAN Z, SHEN C, CHEN H, et al. FCOS:Fully Convolutional One-Stage Object Detection[C]// Proceedings of the IEEE International Conference on Computer Vision.Piscataway:IEEE, 2019:9627-9636.
[10]	HE K, ZHANG X, REN S, et al. Deep Residual Learning for Image Recognition[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2016:770-778.
[11]	LIN T Y, DOLLAR P, GIRSHICK R, et al. Feature Pyramid Networks for Object Detection[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2017:2117-2125.
[12]	LIN T Y, GOYAL P, GIRSHICK R, et al. Focal Loss for Dense Object Detection[C]// Proceedings of the IEEE International Conference on Computer Vision.Piscataway:IEEE, 2017:2980-2988.
[13]	REZATOFIGHI H, TSOI N, GWAK J Y, et al. Generalized Intersection over Union:A Metric and A Loss for Bounding Box Regression[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2019:658-666.
[14]	PANG J, CHEN K, SHI J, et al. Libra R-CNN:Towards Balanced Learning for Object Detection[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2019:821-830.
[15]	ZHU B, WANG J, JIANG Z, et al. AutoAssign:Differentiable Label Assignment for Dense Object Detection (2020)[J/OL].[2020-07-07]. https://arxiv.org/pdf/2007.03496v3.pdf.
[16]	REN S, HE K, GIRSHICK R, et al. Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks[J]. Advances in neural information processing systems, 2015, 28:91-99.
[17]	REDMON J, FARHADI A. Yolov3:An Incremental Improvement (2018)[J/OL].[2018-04-08]. https://arxiv.org/pdf/1804.02767v1.pdf.
[18]	ZHU C, HE Y, SAVVIDES M. Feature Selective Anchor-Free Module for Single-Shot Object Detection[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.Piscataway:IEEE, 2019:840-849.
[19]	KONG T, SUN F, LIU H, et al. FoveaBox:BeyoundAnchor-Based Object Detection[J]. IEEE Transactions on Image Processing, 2020, 29:7389-7398. doi: 10.1109/TIP.2020.3002345
[20]	DI LASCIO R, GRECO A, SAGGESE A, et al. Improving Fire Detection Reliability by A Combination of Videoanalytics[C]// International Conference Image Analysis and Recognition.Heidelberg:Springer, 2014:477-484.

实验环境	环境条件
CPU	Intel Xeon W-2102
GPU	NVIDIA GeForce RTX 2080 Ti
操作系统	Ubuntu 16.0.7
编程语言	Python 3.7
深度学习框架	Pytorch 1.4.0

采样方式	AP	AP₅₀
标准算法	0.542	0.942
中心采样	0.557	0.945
中心分支^[8]	0.557	0.955
文中算法	0.563	0.957

算法	BFP	融合因子	动态采样	AP	AP₅₀	AP_s	AP_m	AP_l
标准算法				0.542	0.942	0.515	0.563	0.561
	√			0.560	0.953	0.521	0.594	0.595
文中算法	√	√		0.569	0.956	0.542	0.596	0.595
	√	√	√	0.583	0.969	0.554	0.611	0.616

算法类型	算法名称	AP	AP₅₀	参数量	速度/(帧每秒)
	Faster RCNN	0.568	0.955	41.12×10⁶	13.7
Anchor Based	YOlOv3	0.525	0.911	61.53×10⁶	34.9
	RetinaNet	0.545	0.956	31.93×10⁶	15.2
	FSAF	0.541	0.954	31.88×10⁶	15.8
Anchor Free	FoveaBox	0.551	0.948	31.88×10⁶	15.9
	FCOS	0.557	0.955	31.84×10⁶	15.7
	文中算法	0.583	0.969	32.05×10⁶	16.2

算法	误报率	漏报率	准确率
文献[20]	13.33	0.00	92.86
文献[1]	11.67	0.00	93.55
文献[2]	4.83	4.53	95.32
文中算法	4.63	0.00	95.86