一种双网融合的分阶段烟雾检测算法

doi:10.19665/j.issn1001-2400.2020.04.019

摘要/Abstract

摘要：

现有的视频烟雾检测方法在复杂场景下检测准确率低,不能准确地框定出视频图像中的烟雾区域。针对此问题,提出一种结合烟雾运动过程和目标检测的分阶段烟雾检测算法。首先,基于烟雾颜色特征改进ViBe算法,提取视频中不断运动的烟雾。然后以YOLO v3模型为基础,在主干网络的残差结构中引入通道注意力机制;使用Focal-loss和GIoU改进损失函数。在烟雾图片数据集测试中,改进后的网络单张图片检测时间为38.4ms,mAP达到约92.13 %,相比原模型提高了约2.19 %。在提取烟雾运动的同时,将同一帧送入网络进行检测;以两者的检测结果对烟雾做综合判别。在公开烟雾视频测试中,该算法的检测率平均达到约98.88%。测试表明,算法对复杂场景适应性强,检测效率高,具有较高的实际应用价值。

关键词: 烟雾检测, ViBe算法, YOLO v3, 注意力机制, 深度学习

Abstract:

Existing video smoke detection methods have a low detection accuracy in complex scenes and cannot detect smoke areas in video frames accurately. In this paper, a phased smoke detection algorithm that combines the smoke movement process and the target detection algorithm is proposed. First, an improved ViBe algorithm based on smoke color features is used to extract the continuously moving smoke in video. Then, the YOLO v3 model is used as the target detection network. The channel attention mechanism is added to the residual structure of its backbone network. Focal-loss and GIoU are utilized to improve the loss function. According to the test of the smoke image data set, the detection time of the improved network on a single picture is 38.4ms and the mAP reaches 92.13%, which is 2.19% higher than that by the original model. While extracting smoke motion, the same frame is sent to the improved YOLO v3 for smoke detection. Finally, comprehensive discrimination is made based on the smoke detection results in stages. Public smoke video test results show that the algorithm has an average detection rate of 98.88%, which proves that the algorithm has a strong adaptability, a high detection efficiency in complex scenes and a high practical application value.

Key words: smoke detection, ViBe algorithm, YOLO v3, attention mechanism, deep learning

中图分类号:

TP391

杜立召,徐岩,张为. 一种双网融合的分阶段烟雾检测算法[J]. 西安电子科技大学学报, 2020, 47(4): 141-148.

DU Lizhao,XU Yan,ZHANG Wei. Phased smoke detection algorithm using dual network fusion[J]. Journal of Xidian University, 2020, 47(4): 141-148.

图/表 9

图1

图2

图3

图4

表1

图5

图6

图7

表2

参考文献 15

[1]	ZEN R I M, WIDYANTO M R, KISWANTO G, et al. Dangerous Smoke Classification Using Mathematical Model of Meaning[J]. Procedia Engineering, 2013,62:963-971. doi: 10.1016/j.proeng.2013.08.149
[2]	KO B C, PARK J O, NAM J Y. Spatiotemporal Bag-of-features for Early Wildfire Smoke Detection[J]. Image and Vision Computing, 2013,31(10):786-795. doi: 10.1016/j.imavis.2013.08.001
[3]	TIAN H, LI W, OGUNBONA P, et al. Smoke Detection in Videos Using Non-redundant Local Binary Pattern-based Features[C]// Proceedings of the 2011 IEEE International Workshop on Multimedia Signal Processing. Washington: IEEE Computer Society, 2011: 6093844.
[4]	TAO C, ZHANG J, WANG P. Smoke Detection Based on Deep Convolutional Neural Networks[C]// Proceedings of the 2016 International Conference on Industrial Informatics-computing Technology, Intelligent Technology, Industrial Information Integration. Piscataway: IEEE, 2016: 150-153.
[5]	YIN Z, WAN B, YUAN F, et al. A Deep Normalization and Convolutional Neural Network for Image Smoke Detection[J]. IEEE Access, 2017,5:18429-18438. doi: 10.1109/ACCESS.2017.2747399
[6]	ZHANG Q X, LIN G H, ZHANG Y M, et al. Wildland Forest Fire Smoke Detection Based on Faster R-CNN Using Synthetic Smoke Images[C]// Proceedings of the 2017 8th International Conference on Fire Science and Fire Protection Engineering. Oxford: Elsevier Ltd, 2018: 441-446.
[7]	REDMON J, FARHADI A. Yolov3: An Incremental Improvement[J/OL]. [2019-12-16].https://arxiv.org/pdf/1804.02767.pdf.
[8]	REDMON J, DIVVALA S, GIRSHICK R, et al. You Only Look Once: Unified, Real-time Object Detection[C]// Proceedings of the 2016 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Washington: IEEE Computer Society, 2016: 779-788.
[9]	REDMON J, FARHADI A. YOLO9000: Better, Faster, Stronger[C]// Proceedings of the 2017 30th IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2017: 6517-6525.
[10]	HE K, ZHANG X, REN S, et al. Deep Residual Learning for Image Recognition[C]// Proceedings of the 2016 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Washington: IEEE Computer Society, 2016: 770-778.
[11]	HU J, SHEN L, SNU G. Squeeze-and-excitation Networks[C]// Proceedings of the 2018 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Washington: IEEE Computer Society, 2018: 7132-7141.
[12]	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 2019 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Washington: IEEE Computer Society, 2019: 658-666.
[13]	赵敏, 张为, 王鑫, 等. 时空背景模型下结合多种纹理特征的烟雾检测[J]. 西安交通大学学报, 2018,52(8):72-78.
	ZHAO Min, ZHANG Wei, WANG Xin, et al. Smoke Detection via Multi-texture Feature Exploration with Spatio-temporal Background Model[J]. Journal of Xi’an Jiaotong University, 2018,52(8):72-78.
[14]	BESBES O, BENAZZA-BENYAHIA A. A Novel Video-based Smoke Detection Method Based on Color Invariants[C]// Proceedings of the 2016 IEEE International Conference on Acoustics, Speech and Signal Processing. Piscataway: IEEE, 2016: 1911-1915.
[15]	汪梓艺, 苏育挺, 刘艳艳, 等. 一种改进DeeplabV3网络的烟雾分割算法[J]. 西安电子科技大学学报, 2019,46(6):52-59.
	WANG Ziyi, SU Yuting, LIU Yanyan, et al. Algorithm for Segmentation of Smoke Using the Improved DeeplabV3 Network[J]. Journal of Xidian University, 2019,46(6):52-59.

模型	mAP/%	时间/ms
SSD	87.56	52.6
YOLOV3	89.94	37.1
Ours1	91.29	38.4
Ours2	92.13	38.4

烟雾视频	文中算法		文献[13]		文献[14]		文献[15]
烟雾视频	R_TPR/ %	R_TNR/ %	R_TPR/ %	R_TNR/ %	R_TPR/ %	R_TNR/ %	R_TPR/ %	R_TNR/ %
sBehindtheFence	98.86	97.60	97.20	96.27	94.72	100	98.26	94.60
sBtFence2	98.40	100	98.17	100	99.08	100	98.20	100
sMoky	98.91	100	99.68	100	86.23	100	98.85	100
sWasteBasket	99.45	99.12	97.18	98.36	99.89	92.60	99.41	100
sWindow	98.80	100	98.10	100	94.30	100	98.40	100
均值	98.88	99.34	98.06	98.83	94.84	98.52	98.56	98.92