基于卷积神经网络和优化GoogleNet架构的监控视频火灾探测  

Fire Detection of Surveillance Video Based on Convolutional Neural Network and Optimized Googlenet Architecture

作　　者：陈钦柱 姚冬 黄松 CHEN Qin-zhu;YAO Dong;HUANG Song(Electric Power Research Institute of Hainan Power Grid Co.,Ltd.,Haikou 570311 China;Key Laboratory of Physical and Chemical Analysis of Hainan Power Grid,Haikou 570311 China)

机构地区：[1]海南电网有限责任公司电力科学研究院,海南海口570311 [2]海南省电网理化分析重点实验室,海南海口570311

出　　处：《自动化技术与应用》2021年第9期124-129,共6页Techniques of Automation and Applications

摘　　要：由于目前主流的神经网络算法通常需要大量的计算时间和较多计算资源,从而限制了其在工程领域中的广泛应用。本文针对火灾视频监控提出了一种具有成本效益的火灾探测CNN架构。该架构基于GoogleNet架构而开发,并在充分考虑目标问题和火灾数据的性质的前提下,具有平衡了计算效率和检测准确性的优势。与其他火焰检测算法相比,该检测方法具有更合理的计算复杂性和对预期火灾检测问题的较好适用性。实验结果证明了相对于其他火灾检测方法,所提出的检测方法具有更加出色的检测准确度和十分优异的抗干扰能力,十分适合用于集成在视频监视系统实现对山林火灾的早期预警。Because current mainstream neural network algorithms usually require a large amount of computation time and more computing resources,it limits its wide application in engineering.This paper proposes a cost-effective CNN architecture for fire video surveillance.The architecture is developed based on the GoogleNet architecture and balances computational efficiency and detection accuracy with full consideration of the target problem and the nature of the fire data.Compared with other flame detection algorithms,this detection method has more reasonable computational complexity and applicability to expected fire detection problems.The experimental results show that the proposed detection method has better detection accuracy and excellent anti-interference ability than other fire detection methods,and is very suitable for integration in the video surveillance system to achieve early warning of forest fires.

关 键 词：火灾探测 图像分类 深度学习 视频分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]