摘要
一般火焰检测方法由于对复杂场景的应变能力较差,因此检测率较低。文中提出了一种基于改进的YOLOv2网络的深度学习火焰检测方法,来自动提取火焰特征;同时,针对特征提取过程中信息丢失的问题,采用聚类选取候选框,以多尺度特征融合的方法融合高层与浅层特征信息,进一步提高了模型的检测率。在Bilkent大学火焰视频数据集上的实验结果表明,该方法的平均正检率达到了98.8%,检测速率达到40帧/s,具有较强的鲁棒性和实时性。
It is difficult for general flame detection methods to adapt to complex scenes,so the detection rates is low.This paper proposed a deep learning flame detection method based on an improved YOLOv2 network to extract the flame features automatically.In order to avoid the information loss in the feature extraction process,the selected anchor box by clustering is suggested and multi-scale feature fusion method is used to fuse high-level and shallow feature information,to further improve the detection rate of the model.Experimental results on the Bilkent University flame video dataset show that the average true inspection rate of the proposed method is 98.8%,and the detection rate is 40 frames/s,so its robustness and real-time performance are strong.
引文
[1] DIMITROPOULOS K,BARMPOUTIS P,GRAMMALIDIS N.Spatio-Temporal Flame Modeling and Dynamic Texture Analysis for Automatic Video-Based Fire Detection[J].IEEE Tran-sactions on Circuits & Systems for Video Technology,2015,25(2):339-351.
[2] CELIK T,DEMIREL H,OZKARAMANLI H.Automatic fire detection in video sequences[C]//European Signal Processing Conference.IEEE,2015:1-5.
[3] POOBALAN K,LIEW S C.Fire detection algorithm using ima- ge processing techniques[C]//International Conference on Artificial Intelligence and Computer Science.2015.
[4] GENG Q T,YU F H,ZHAO H W,et al.New algorithm of flame detection based on color features[J].Journal of Jilin University(Engineering and Technology Edition),2014,44(6):1787-1792.(in Chinese)耿庆田,于繁华,赵宏伟,等.基于颜色特征的火焰检测新算法[J].吉林大学学报(工学版),2014,44(6):1787-1792.
[5] FRIZZI S,KAABI R,BOUCHOUICHA M,et al.Convolutional neural network for video fire and smoke detection[C]//Confe-rence of the IEEE Industrial Electronics Society.IEEE,2016:877-882.
[6] MUHAMMAD K,AHMAD J,MEHMOOD I,et al.Convolu- tionalNeural Networks based Fire Detection in Surveillance Vi-deos[J].IEEE Access,2018,PP(99):1.
[7] ZHANG H,WANG K F,WANG F Y.Advances and Perspectives on Applications of Deep Learning in Visual Object Detection[J].Acta Automatic Sinica,2017,43(8):1289-1305.(in Chinese)张慧,王坤峰,王飞跃.深度学习在目标视觉检测中的应用进展与展望[J].自动化学报,2017,43(8):1289-1305.
[8] REDMON J,FARHADI A.YOLO9000:Better,faster,stronger[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Washington:IEEE Computer Society,2017:6517-6525.
[9] REN S,GIRSHICK R,GIRSHICK R,et al.Faster R-CNN:Towards Real-Time Object Detection with Region Proposal Networks[J].IEEE Transactions on Pattern Analysis & Machine Intelligence,2017,39(6):1137-1149.
[10] HE K,ZHANG X,REN S,et al.Identity Mappings in Deep Residual Networks[C]//Computer Vision-ECCV 2016.Springer International Publishing,2016:630-645.
[11] PINHEIRO P O,LIN T Y,COLLOBERT R,et al.Learning to refine object segments[C]//Lecture Notes in Computer Scien-ce.Heidelberg.Springer,2016:75-91.
[12] LIU W,ANGUELOV D,ERHAN D,et al.SSD:Single Shot MultiBox Detector[C]//European Conference on Computer Vision.Springer,Cham,2016:21-37.
[13] HOWARD A G,ZHU M,CHEN B,et al.MobileNets:Efficient Convolutional Neural Networks for Mobile Vision Applications[J].arXiv preprint arXiv:1704.04861,2017.