基于烟雾区域和轻量化模型的视频烟雾检测  

Video Smoke Detection based on Smoke Area and Lightweight Model

作　　者：蒲建飞 魏维 吴帝勇 程鹏 袁定胜 PU Jianfei;WEI Wei;WU Diyong;CHENG Peng;YUAN Dingshen(College of Software Engineering,Chengdu University of Information Technology,Chengdu 610225,China)

机构地区：[1]成都信息工程大学软件工程学院,四川成都610225

出　　处：《成都信息工程大学学报》2023年第3期281-290,共10页Journal of Chengdu University of Information Technology

基　　金：四川省科技厅重点科研资助项目(2021YFG0299)。

摘　　要：烟雾是早期火灾发生的典型特征,针对烟雾的智能检测能有效降低森林火灾造成的破坏。为了对监控视频中烟雾的及早检测,提出一种基于烟雾区域及轻量化模型的烟雾检测算法。首先通过残差帧堆叠获取视频中的运动区域,然后再利用自适应暗通道掩码对运动区域进一步筛选获得疑似烟雾块。在此基础上,又设计了一个轻量化的神经网络模型用于烟雾识别,模型利用卷积局部感知的特性提取烟雾的浅层特征,而在网络深层则将卷积和self-attention相结合,通过比较全局相似度,在浅层特征图的基础上获取烟雾的全局信息。实验结果表明,算法具有较强的鲁棒性,无论是远距离烟雾还是近距离烟雾均有良好的检测效果。Smoke is a typical feature of early fires,and intelligent detection of smoke can effectively reduce the damage caused by forest fires.In order to achieve early detection of smoke in surveillance video,this paper proposes a smoke detection algorithm based on the smoke area and lightweight model.The algorithm first obtains the motion regions in the video by stacking the residual frames,and then uses the adaptive dark channel mask to further screen the motion regions to obtain suspected smoke blocks.On this basis,this paper proposes a lightweight neural network model for smoke recognition.This model uses the characteristics of convolutional local perception to extract the shallow features of smoke,and in the deep layer of the network,convolution and self-attention are related.The global information of smoke is obtained on the basis of shallow feature maps by comparing the global similarity.The experimental results show that the algorithm in this paper has strong robustness,and it has a good detection effect for long-distance smoke and short-distance smoke.

关 键 词：早期烟雾 暗通道 自注意力机制 卷积神经网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]