摘要
针对多种环境下早期火灾烟雾小导致的目标检测精度不足,以及特征表达能力不充分等问题,提出了一种基于YOLOv5s的火灾烟雾检测改进算法。首先,对火灾烟雾图像数据集的结构进行了详细分析;然后,根据数据集结构特点,在主干网引入了通道注意力机制与空间注意力机制;最后,增加了小目标检测层,以更加有效地关注小目标的检测。实验结果表明,改进后的YOLOv5s算法精确率达到了89.9%,其平均准确率均值PmA相较于原YOLOv5s的提高了5.0%。该模型在火灾烟雾检测方面表现出优异效果,对火灾烟雾的早期预警具有指导意义。
A fire smoke detection algorithm based on YOLOv5s is proposed to solve the problems of insufficient accuracy and insufficient feature expression ability of small objects detection in early fire smoke in various environments.Firstly,a detailed analysis of the structure of the fire smoke image dataset is conducted.Then,according to the characteristics of the dataset structure,channel attention mechanism and spatial attention mechanism are introduced into the backbone network.Finally,a small object detection layer is added to pay more attention to the detection of small objects.The experimental results show that the accuracy of the improved YOLOv5s algorithm reaches 89.9%,and its average accuracy PmA is increased by 5.0% compared with the original YOLOv5s.The model shows excellent results in fire smoke detection and has guiding significance for early warning of fire smoke.
作者
姚芳
YAO Fang(College of Mechanical and Automotive Engineering,Chuzhou Polytechnic,Chuzhou,Anhui 239000,China)
出处
《湖南城市学院学报(自然科学版)》
2025年第1期60-66,共7页
Journal of Hunan City University:Natural Science
基金
安徽省优秀青年教师培育项目(YQYB2023161)。
