摘要
提出了一种改进的基于YOLOv7的火灾现场行人检测算法。首先,利用自动色阶算法对火灾现场图像进行预处理;然后,采用HorBlock与CSPNet构造HorBc模块,改进YOLOv7网络结构,加强特征提取能力;同时融合CBAM注意力机制,增加行人特征区域学习权重。实验结果表明:在收集的火灾现场行人数据集上平均精度为97.1%,召回率为95.6%,精确率达到了97.6%;相比原始YOLOv7算法,平均精度提升了1.5%,召回率提升了2.4%,精确率提升了1.8%,在实时性上达到了36.7 fps,满足实时性要求。
An improved fire scenes pedestrian detection algorithm based on YOLOv7 is proposed.Firstly,automatic color scale algorithm is used to preprocess the fire scene image.Then,in order to enhance the feature extraction capability,YOLOv7 network structure is improved by using HorBc module,which is constructed by combining HorBlock and CSPNet networks.At the same time,fuse CBAM attention mechanism,increase the weight of pedestrian feature area learning.Experimental results show that the average precision(AP)of 97.1%,the recall rate of 95.6%,and the precision of 97.6%are achieved on the collected pedestrian dataset in the fire scene,compared to the original YOLOv7 algorithm,the AP is increased by 1.5%,the recall rate is increased by 2.4%,and the precision is increased by 1.8%.In real-time performance,36.7 fps is achieved,which meets the real-time requirement.
作者
赵伟
沈乐
徐凯宏
ZHAO Wei;SHEN Le;XU Kaihong(College of Computer and Control Engineering,Northeast Forestry University,Harbin 150040,China)
出处
《传感器与微系统》
CSCD
北大核心
2023年第7期165-168,共4页
Transducer and Microsystem Technologies
基金
国家自然科学基金资助项目(61975028)
黑龙江省重点研发计划资助项目(GZ20210017,GZ20210018)。
关键词
行人检测
YOLOv7
自动色阶
HorBc
卷积块注意模块
pedestrian detection
YOLOv7
automatic color scale
HorBc
convolutional block attention module(CBAM)