摘要
本文采用合成图像的Faster R-CNN对森林火灾烟雾进行检测,避免了传统视频烟雾检测方法中复杂的人工特征提取过程.合成烟雾图像是将真实或模拟烟雾插入到森林背景中,解决了训练数据缺乏的问题.将真实合成烟雾和模拟合成烟雾分别训练后的模型放在由真实火焰烟雾图像组成的数据集中测试,测试结果表明,模拟烟雾是更好的选择,模型对薄烟不敏感.通过改进森林火灾烟雾图像的合成过程或者将这个解决方案扩展到视频序列中,可以进一步提高它的性能.
In this paper, Faster R-CNN was used to detect forest fire smoke to avoid the complex manually feature extraction process in traditional video smoke detection methods. Synthetic smoke images are produced by inserting real smoke or simulative smoke into forest background to solve the lack of training data. The models trained by the two kinds of synthetic images respectively are tested in dataset consisting of real fire smoke images. The results show that simulative smoke is the better choice and the model is insensitive to thin smoke. It may be possible to further boost the performance by improving the synthetic process of forest fire smoke images or extending this solution to video sequences.
引文
[1]Genovese A,Labati R D,Piuri V.Wildfire smoke detection using computational intelligence techniques[C].IEEE International Conference on Computational Intelligence for Measurement Systems Applications,2011.
[2]Yuan F.Video-based smoke detection with histogram sequence of lbp and lbpv pyramids[J].Fire Safety Journal,2011,46(3):132-139.
[3]Toreyin B U,Dedeoglu Y,Cetin A E.Contour based smoke detection in video using wavelets[C].European Signal Processing Conference,2006.
[4]张永明,于春雨,袁非牛.一种基于光流法的视频火灾烟雾图像识别方法[P].中国发明专刊,2008100231477,2008-07-15.
[5]贾阳,林高华,王进军.基于显著性检测和高斯混合模型的早期视频烟雾分割算法[J].计算机工程,2016,42(2):206-209.
[6]Hohberg S P.Wildfire smoke detection using convolutional neural networks[C].Freie Universitt Berlin,2015.
[7]Frizzi S,Kaabi R,Bouchouicha M.Convolutional neural network for video fire and smoke detection[C].Conference of The IEEE Industrial Electronics Society,2016:877-882.
[8]傅天驹,郑嫦娥,田野.复杂背景下基于深度卷积神经网络的森林火灾识别[J].计算机与现代化,2016,(3):52-57.
[9]Genovese A,Labati R D,Piuri V.Virtual environment for synthetic smoke clouds generation[C].International Conference on Virtual Environments Human-computer Interfaces Measurement Systems,2011:1-6.
[10]Collins R T,Lipton A J,Kanade T.A system for video surveillance and monitoring[C].The proceeding of American nuclear society eighth international topical meeting on robotics and remote systems,Pittsburgh,PA,1999.
[11]陈晓娟,卜乐平,李其修.基于图像处理的明火火灾探测研究[J].海军工程大学学报,2007,19(3):6-11.
[12]Ren S,He K,Girshick R.Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks[C].International Conference on Neural Information Processing Systems,2015.
[13]曾国荪.改善神经网络反向传播算法的训练时间[J].小型微型计算机系统,1996,(11):69-72.
[14]王功鹏,段萌,牛常勇.基于卷积神经网络的随机梯度下降算法[J].计算机工程与设计,2018,(2):441-445.
[15]Zeiler,Matthew D,Fergus.Visualizing and Understanding Convolutional Networks[J].European Conference on Computer Vision,2014,8689:818-833.