Journal of Computer Applications ›› 2020, Vol. 40 ›› Issue (5): 1465-1469.DOI: 10.11772/j.issn.1001-9081.2019091583

• Virtual reality and multimedia computing • Previous Articles     Next Articles

Smoke recognition method based on dense convolutional neural network

CHENG Guangtao1, GONG Jiachang2, LI Jian1   

  1. 1.Department of Research and Development, National Center for Fire Engineering Technology, Tianjin 300381, China
    2.Department of Audio-Visual Information Detection Technology, Criminal Investigation Police University of China,ShenyangLiaoning 110854, China
  • Received:2019-09-17 Revised:2019-10-25 Online:2020-05-10 Published:2020-05-15
  • Contact: CHENG Guangtao, born in 1983, Ph. D., engineer. His research interests include image processing, pattern recognition, deep learning.
  • About author:CHENG Guangtao, born in 1983, Ph. D., engineer. His research interests include image processing, pattern recognition, deep learning.GONG Jiachang, born in 1983, Ph. D., lecturer. His research interests include image processing, pattern recognition.LI Jian, born in 1992, M. S., engineer. Her research interests include data mining, image processing.
  • Supported by:

    This work is partially supported by the Program of Tianjin Fire Research Institute of Ministry of Emergency Management (2018SJ20).


程广涛1, 巩家昌2, 李建1   

  1. 1.国家消防工程技术研究中心 研发部,天津 300381
    2.中国刑事警察学院 声像资料检测技术系,沈阳 110854
  • 通讯作者: 程广涛(1983—)
  • 作者简介:程广涛(1983—),男,黑龙江大庆人,工程师,博士,主要研究方向:图像处理、模式识别、深度学习; 巩家昌(1983—),男,山东临沂人,讲师,博士,主要研究方向:图像处理、模式识别; 李建(1992—),女,河北沧州人,工程师,硕士,主要研究方向:数据挖掘、图像处理。
  • 基金资助:



To address the poor robustness of the extracted image features in traditional smoke detection methods, a smoke recognition method based on Dense convolution neural Network (DenseNet) was proposed. Firstly, the dense network blocks were constructed by applying convolution operation and feature map fusion, and the dense connection mechanism was designed between the convolution layers, so as to promote the information circulation and feature reuse in the dense network block structure. Secondly, the DenseNet was designed by stacking the designed dense network blocks for smoke recognition, saving the computing resources and enhancing the expression ability of smoke image features. Finally, aiming at the problem of small smoke image data size, data augmentation technology was adopted to further improve the recognition ability of the training model. Experiments were carried out on public smoke datasets. The experimental results illustrate that the proposed method achieves high accuracy of 96.20% and 96.81% on two test sets respectively with only 0.44 MB model size.

Key words: smoke recognition, dense connection, Convolutional Neural Network (CNN), deep learning, data augmentation


针对传统烟雾检测方法中提取的图像特征鲁棒性较差的问题,提出了基于稠密卷积神经网络(DenseNet)的烟雾识别方法。首先,利用卷积操作和特征图融合构建稠密网络块,在卷积层之间设计稠密连接机制,以增强稠密网络块结构内的信息流通和特征重利用;然后,将已构建的稠密网络块叠加成稠密卷积神经网络用于烟雾识别,节省计算资源的同时提升对烟雾图像特征的表达能力;最后,针对烟雾图像数据量较小的问题,采取数据增强技术进一步改善训练模型的识别能力。在公开烟雾数据集上对提出的方法进行实验验证,实验结果表明,所提方法的模型大小只有0.44 MB,在两个测试集上的准确率分别为96.20%和96.81%。

关键词: 烟雾识别, 稠密连接, 卷积神经网络, 深度学习, 数据增强

CLC Number: