摘要
针对现有烟火检测算法存在的漏检和误检问题,提出一种基于高效全局上下文网络(EGC-Net)的轻量级烟火检测新算法。该算法以轻量级目标检测网络YOLOX为基础网络,将改进的EGC-Net嵌入到YOLOX的主干特征提取网络与特征金字塔网络之间。EGC-Net由上下文建模、特征转换和特征融合3阶段结构组成,用于获得图像的全局上下文信息,建模烟火目标与其背景信息的远程依赖关系,并结合通道注意力机制学习更具判别力的视觉特征用于烟火检测。实验结果表明,本文提出的EGC-YOLOX烟火检测算法的图像级召回率为95.56%,图像级误报率为4.75%,均优于对比的其他典型轻量级算法,且速度满足实时检测的要求。该算法可在安防和消防领域推广,用于实时火灾监控和预警管理。
Aiming at the problems of missing and false detection in existing smoke and fire detection algorithms,a new lightweight algorithm based on Effective Global Context Network(EGC-Net)is proposed. It takes the lightweight object detection network YOLOX as the basic network,and embeds an improved EGC-Net between the backbone feature extraction network and feature pyramid network of YOLOX. EGC-Net is composed of a three-stage structure of context modeling,feature transformation and feature fusion,which is used to obtain the global context information of image,model the long-range dependency of smoke or fire objects and its background,and learn more discriminative visual features by combining the channel attention mechanism for smoke and fire detection. Experimental results indicate that the image-level recall rate of the proposed smoke and fire detection algorithm EGC-YOLOX is 95. 56%,and the image-level false alarm rate is 4. 75%,both of which are superior to the compared typical lightweight algorithms,and the speed also meets the requirements of real-time detection. The proposed algorithm can be promoted and applied to the field of security and fire protection for real-time fire monitoring and early warning management.
作者
魏伦胜
徐望明
张景元
陈彬
WEI Lun-sheng;XU Wang-ming;ZHANG Jing-yuan;CHEN Bin(School of Information Science and Engineering,Wuhan University of Science and Technology,Wuhan 430081,China;Engineering Research Center for Metallurgical Automation and Detecting Technology of Ministry of Education,Wuhan University of Science and Technology,Wuhan 430081,China)
出处
《液晶与显示》
CAS
CSCD
北大核心
2023年第1期118-127,共10页
Chinese Journal of Liquid Crystals and Displays
基金
国家自然科学基金(No.51805386)
教育部冶金自动化与检测技术工程研究中心开放课题(No.MADTOF2021B02)
湖北省教育厅科研计划(No.D20191104)。
