南京大学学报(自然科学版) ›› 2023, Vol. 59 ›› Issue (2): 295–301.doi: 10.13232/j.cnki.jnju.2023.02.012

• • 上一篇    下一篇

基于CXANet⁃YOLO的火焰检测方法

卞苏阳1, 严云洋1,2(), 龚成张1, 冷志超1, 祝巧巧1   

  1. 1.淮阴工学院计算机与软件工程学院,淮安,223003
    2.江苏海洋大学计算机工程学院,连云港,222005
  • 收稿日期:2022-11-14 出版日期:2023-03-31 发布日期:2023-04-07
  • 通讯作者: 严云洋 E-mail:yunyang@hyit.edu.cn
  • 基金资助:
    国家自然科学基金(61402192);江苏省“六大人才高峰”项目(2013DZXX?023);江苏省“青蓝工程”,淮安市“533英才工程”

Flame detection based on CXANet⁃YOLO

Suyang Bian1, Yunyang Yan1,2(), Chengzhang Gong1, Zhichao Leng1, Qiaoqiao Zhu1   

  1. 1.Faculty of Computer & Software Engineering,Huaiyin Institute of Technology,Huai'an,223003,China
    2.School of Computer Engineering,Jiangsu Ocean University,Lianyungang,222005,China
  • Received:2022-11-14 Online:2023-03-31 Published:2023-04-07
  • Contact: Yunyang Yan E-mail:yunyang@hyit.edu.cn

摘要:

快速准确的火焰检测对于降低火灾危害具有重要意义,为了加强模型的火焰特征提取能力以及解决特征图尺寸不平衡的问题,利用XSepConv (Extremely Separated Convolution)、大卷积核、Mish激活函数等构建CXANet?block (Convolution Extremely Attention Network)作为YOLOv5的骨干网络,引入CBAM (Convolution Block Attention Module)注意力机制,提出一种基于CXANet?YOLO的火焰检测方法,通过增加通道注意力和空间注意力来提高检测性能.在自建火焰数据集上进行训练,提升模型的鲁棒性和泛化能力.实验结果表明,CXANet?YOLO模型比基准模型YOLOv5在火焰检测上具有更高的检测精度和检测速度,准确率提高了8.2%,检测速度每秒提升25帧.

关键词: 深度学习, 火焰检测, 注意力机制, YOLOv5

Abstract:

It is great significant for reducing fire hazards based fast and accurate flame detection. In order to strengthen the flame feature extraction capability of the model and solve the problem of feature map size imbalance,CXANet?block (Convolution Extremely Attention Network) is built as the backbone network of YOLOv5 with XSepConv (Extremely Separated Convolution),large convolution kernel,Mish activation function,etc. CBAM (Convolution Block Attention Module) is introduced,and a CXANet?YOLO?based flame detection method is proposed to improve the detection performance by increasing channel attention and spatial attention. It is trained on the self?built flame dataset to improve the robustness and generalization of the model. Experimental results show that the CXANet?YOLO model has higher detection accuracy and detection speed than the benchmark model YOLOv5 in flame detection. The accuracy rate is increased by 8.2%,and the detection speed is added by 25 frames per second.

Key words: deep learning, flame detection, attentional mechanism, YOLOv5

中图分类号: 

  • TP391

图1

YOLOv5的网络结构"

图2

C3模块的结构"

图3

SPPF模块的结构"

图4

XSepConv的基础结构"

图5

通道注意力机制"

图6

空间注意力机制"

图7

CXANet的基础结构"

图8

训练集火焰数据的范例"

图9

测试集火焰数据的范例"

表1

消融实验的结果"

ModelsPAPFPS
YOLOv590.6%66.9%14
YOLOv5_192.5%68.0%15
YOLOv5_297.4%67.5%35
YOLOv5_398.8%69.2%39

表2

模型性能的对比"

ModelsPAPFPS
YOLOv477.82%63.11%28
YOLOv590.73%66.91%14
YOLOx[16]84.23%73.30%21
Faster⁃RCNN[17]34.41%68.09%8
CenterNet[18]77.95%54.40%39
RetinaNet[19]67.69%59.27%23
CXANet⁃YOLO98.80%69.20%39
1 Emmy P C, Vinsley S S, Suresh S. Efficient flame detection based on static and dynamic texture analysis in forest fire detection. Fire Technology201854(1):255-288.
2 张慧珍,严云洋,刘以安,等. 基于超像素分割与闪频特征判别的视频火焰检测. 数据采集与处理201833(3):512-520.
Zhang H Z, Yan Y Y, Liu Y A,et al. Video flame detection based on super pixel segmentation and flash frequency feature discrimination. Journal of Data Acquisition and Processing201833(3):512-520.
3 Wu X H, Lu X B, Leung H. A video based fire smoke detection using robust AdaBoost. Sensors201818(11):3780.
4 Zhong Z, Wang M J, Shi Y K,et al. A convolutional neural network?based flame detection method in video sequence. Signal,Image and Video Processing201812(8):1619-1627.
5 邓军,姚涵文,王伟峰,等. 基于优化InceptionV1的视频火焰超像素检测方法. 激光与光电子学进展202158(2):0210004.
Deng J, Yao H W, Wang W F,et al. Detection method for video flame super?pixel based on optimized InceptionV1. Laser & Opto?electronics Progress202158(2):0210004.
6 Qin Y Y, Cao J T, Ji X F. Fire detection method based on depthwise separable convolution and YOLOv3. International Journal of Automation and Computing202118(2):300-310.
7 吕璐,程虎,朱鸿泰,等. 基于深度学习的目标检测研究与应用综述. 电子与封装202222(1):010307.
Lü L, Cheng H, Zhu H T,et al. Progress of research and application of object detection based on deep learning. Electronics & Packaging202222(1):010307.
8 Wang Z, Wu L, Li T,et al. A smoke detection model based on improved YOLOv5. Mathematics202210(7):1190.
9 Bochkovskiy A, Wang C Y, Liao H Y M. YOLOv4:Optimal speed and accuracy of object detection. Computer Vision and Pattern Recognition202017(9):198-215.
10 Chen J R, Lu Z Q, Liao Q M. XSepConv:Extremely separated convolution for efficient deep networks with large kernels∥Proceedings of SPIE 11878,the 13th International Conference on Digital Image Processing. Singapore,Singapore:SPIE,2021,Article No.65.
11 Woo S, Park J, Lee J Y,et al. CBAM:Convolutional block attention module∥Proceedings of the 15th European Conference on Computer Vision. Springer Berlin Heidelberg,2018:3-19.
12 Fu H X, Song G Q, Wang Y C. Improved YOLOv4 marine target detection combined with CBAM. Symmetry202113(4):623.
13 Zhang Z L, Jin Q Y, Liu Z N,et al. Video?based fire smoke detection using temporal?spatial saliency features. Procedia Computer Science2022(198):493-498.
14 Liu Z, Mao H Z, Wu C Y,et al. A ConvNet for the 2020s∥Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans,LA,USA:IEEE,2022:11976-11986.
15 Ding X H, Zhang X Y, Han J G,et al. Scaling up your kernels to 31x31:Revisiting large kernel design in CNNs∥Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New Orleans,LA,USA:IEEE,2022:11963-11975.
16 Ge Z, Liu S T, Wang F,et al. YOLOX:Exceeding YOLO series in 2021. 2021,arXiv:.
17 Ren S Q, He K M, Girshick R,et al. Faster R?CNN:Towards real?time object detection with region proposal networks. IEEE Transactions on Pattern Analysis and Machine Intelligence201739(6):1137-1149.
18 Duan K W, Bai S, Xie L X,et al. CenterNet:Keypoint triplets for object detection∥Proceedings of 2019 IEEE/CVF International Conference on Computer Vision. Seoul,Korea (South):IEEE,2019:6569-6578.
19 Lin T Y, Goyal P, Girshick R,et al. Focal loss for dense object detection∥Proceedings of 2017 IEEE International Conference on Computer Vision. Venice,Italy:IEEE,2017:2980-2988.
[1] 冯海, 马甲林, 许林杰, 杨宇, 谢乾. 融合标签嵌入和知识感知的多标签文本分类方法[J]. 南京大学学报(自然科学版), 2023, 59(2): 273-281.
[2] 方巍, 李佳欣, 陆文赫. 基于3D卷积和自注意力机制的卫星云图预测研究[J]. 南京大学学报(自然科学版), 2023, 59(1): 155-164.
[3] 林灏昶, 秦云川, 蔡宇辉, 李肯立, 唐卓. 基于目标检测的图形用户界面控件识别方法[J]. 南京大学学报(自然科学版), 2022, 58(6): 1012-1019.
[4] 马学森, 马吉, 蒋功辉, 许雪梅, 周天保. 基于注意力机制和多尺度特征融合的绝缘子缺陷检测方法[J]. 南京大学学报(自然科学版), 2022, 58(6): 1020-1029.
[5] 周佳倩, 林培光, 李庆涛, 王基厚, 刘利达. MDDE:一种基于投资组合的金融市场趋势分析方法[J]. 南京大学学报(自然科学版), 2022, 58(5): 876-883.
[6] 蔡国永, 兰天. 基于多头注意力和词共现关系的方面级情感分析[J]. 南京大学学报(自然科学版), 2022, 58(5): 884-893.
[7] 罗思涵, 杨燕. 一种基于深度学习和元学习的出行时间预测方法[J]. 南京大学学报(自然科学版), 2022, 58(4): 561-569.
[8] 苏雅茜, 崔超然, 曲浩. 基于自注意力移动平均线的时间序列预测[J]. 南京大学学报(自然科学版), 2022, 58(4): 649-657.
[9] 杜渊洋, 邓成伟, 张建. 基于深度卷积神经网络的RNA三维结构打分函数[J]. 南京大学学报(自然科学版), 2022, 58(3): 369-376.
[10] 陈轶洲, 刘旭生, 孙林檀, 李文中, 方立兵, 陆桑璐. 基于图神经网络的社交网络影响力预测算法[J]. 南京大学学报(自然科学版), 2022, 58(3): 386-397.
[11] 张玮, 赵永虹, 邱桃荣. 基于注意力机制和深度学习的运动想象脑电信号分类方法[J]. 南京大学学报(自然科学版), 2022, 58(1): 29-37.
[12] 孟浩, 刘强. 基于FPGA的卷积神经网络训练加速器设计[J]. 南京大学学报(自然科学版), 2021, 57(6): 1075-1082.
[13] 陈磊, 孙权森, 王凡海. 基于深度对抗网络和局部模糊探测的目标运动去模糊[J]. 南京大学学报(自然科学版), 2021, 57(5): 735-749.
[14] 倪斌, 陆晓蕾, 童逸琦, 马涛, 曾志贤. 胶囊神经网络在期刊文本分类中的应用[J]. 南京大学学报(自然科学版), 2021, 57(5): 750-756.
[15] 杨静, 赵文仓, 徐越, 冯旸赫, 黄金才. 一种基于少样本数据的在线主动学习与分类方法[J]. 南京大学学报(自然科学版), 2021, 57(5): 757-766.
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!