摘要
火焰前期一般多为小目标,但一般的火焰检测方法对于小目标的检测能力较差.为检测早期火焰,提高火灾预防能力,提出了一种融合多级特征的视频火焰检测方法,针对下采样分辨率变小导致丢失目标的问题,引入了反卷积模块,并融合深层具有较强语义信息的特征和浅层具有较强细节信息的特征,从而有效提高了火焰的检测率.所提算法在Bilkent大学火灾数据库VisiFire数据集上进行的实验表明,mAP相较于Yolov2提高了10.0%,与多种经典的深度学习算法模型相比,检测率更高.
In the early stage,the flame is usually small target,but the general flame detection method has poor detection ability for small target.To detect early flame and improve fire prevention ability,a video flame detection method with multi-stage features is proposed.Aiming at the problem of missing target caused by the lower sampling resolution,a deconvolution module is introduced,and the features with strong semantic information in the deep layer and strong detail information in the shallow layer are combined to effectively improve the detection rate of flame.The experiment of the proposed algorithm on the Fire database VisiFire data set of Bilkent University showed that mAP improved by 10.0%compared with the Yolov2,and the detection rate was higher compared with many classic deep learning algorithm models.
作者
严忱
严云洋
高尚兵
朱全银
Yan Chen;Yan Yunyang;Gao Shangbing;Zhu Quanyin(Faculty of Computer&Software Engineering,Huaiyin Institute of Technology,Huai’an 223003,China;School of Computer Engineering,Jiangsu Ocean University,Lianyungang 225005,China)
出处
《南京师大学报（自然科学版）》
CAS
CSCD
北大核心
2021年第3期131-136,共6页
Journal of Nanjing Normal University(Natural Science Edition)
基金
国家自然科学基金项目(61402192)
江苏省“六大人才高峰”项目(2013DZXX-023)
江苏省高校自然科学基金重大项目(18KJA52001)。
