一种基于MobileNet的火灾烟雾检测方法  

Fire and Smoke Detection Method Based on MobileNet

作　　者：徐梓涵 刘军[1,2] 张苏沛[1,2] 肖澳文 杜壮 XU Zihan;LIUJun;ZHANG Supei;XIAO Aowen;DU Zhuang(Hubei Key Laboratory of Intelligent Robot(Wuhan Institute of Technology),Wuhan 430205,China;School of Computer Science&Engineering,Wuhan Institute of Technology,Wuhan 430205,China)

机构地区：[1]智能机器人湖北省重点实验室(武汉工程大学),湖北武汉430205 [2]武汉工程大学计算机科学与工程学院,湖北武汉430205

出　　处：《武汉工程大学学报》2019年第6期580-585,共6页Journal of Wuhan Institute of Technology

基　　金：国家自然科学基金（61172150,61803286）;智能机器人湖北省重点实验室开放基金（HBIR 201802）;武汉工程大学第十届研究生教育创新基金（CX2018197,CX2018200,CX2018212)

摘　　要：提出了基于图像序列的火灾烟雾检测方法。首先使用K-近邻(K-NN)背景减除器预测前景区域,对该区域进行形态学操作后得到可能出现火焰或烟雾的区域。其次,使用轻量神经网络MobileNet对火焰和烟雾进行分类。该模型具有流线型架构,同时采用depthwise separate convolution,使得该模型可以运行在嵌入式设备和普通PC机上。实验首先在数据集上完成分类模型训练,使用多种标准进行评估。结果表明:该方法能够在嵌入式设备等计算能力有限的设备上实现火灾烟雾检测。与其他模型相比,该方法在没有明显损失准确度的情况下大幅提高了检测效率。The detection method based on image sequences was proposed in the paper.Firstly,the foreground area was extracted by a K-nearest neighbor classifier.Then,the potential area of fire and smoke was recognized based on morphological operations.Finally,a lightweight neural network MobileNet was used to classify fire and smoke.MobileNet has a streamlined architecture and employs the depth wise separate convolution,which makes it possible to run on both personal computer and embedded devices.In experiments,a classifier was trained on a dataset,and was evaluated according to multiple metrics.The results show that the proposed method is able to detect fire and smoke on embedded devices,and improves the detection efficiency without significant loss of accuracy in comparison with other methods.

关 键 词：火灾烟雾 检测 MobileNet K-近邻算法 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]