首页|期刊导航|沈阳航空航天大学学报|基于YOLOv8n-MAAI的火灾图像检测

基于YOLOv8n-MAAI的火灾图像检测OA

Fire image detection based on YOLOv8n-MAAI

中文摘要英文摘要

为提高复杂环境中火灾检测的准确率和效率,对YOLOv8n模型进行了多项改进,提出一种基于YOLOv8n-MAAI的火灾检测算法.首先,在主干网络YOLOv8n中引入了多头自注意力(multi-head self-attention,MHSA)机制,提高模型处理复杂场景的能力,进而提升了检测精度.其次,将YOLOv9模型中的ADown模块集成到YOLOv8架构中,通过轻量化设计策略减少模型参数数量,降低了计算复杂度.然后,引入自适应空间特征融合(adaptively spa-tial feature fusion,ASFF)机制改进检测头,形成了新的Detect_ASFF检测头,有效过滤冲突信息,提升了特征的尺度不变性.最后,采用Inner-MPDIoU损失函数优化训练过程,增强了回归结果的准确性并加速了模型的收敛速度.试验结果表明,改进后的YOLOv8n-MAAI算法模型检测精度为92.4%,召回率为93.6%,平均精度均值达到94.3%,优于原始的 YOLOv8n、YOLOv5s 和 YOLOv3-tiny等其他流行的检测模型.这些改进为火灾探测提供了更准确、更可靠的解决方案,为火灾预防和控制技术的进步提供了支持.

In order to improve the accuracy and efficiency of fire monitoring in complex enviroments,an improved YOLOv8n model named YOLOv8n-MAAI for fire detection was proposed.Firstly,by incorporating a multi-head self-attention(MHSA)mechanism into the backbone network,the model's ability to handle complex scenarios was enhanced,improving detection accuracy.Furthermore,the ADown module from YOLOv9 was integrated into the YOLOv8 architecture,reducing computational complexity through a lightweight design.The adaptive spatial feature fusion(ASFF)mechanism was introduced to form a new Detect_ASFF detection header,filtering conflicting information and improving feature scale invariance.Finally,the Inner-MPDIoU loss function was employed to optimize the training process,enhancing regression accuracy and accelerating model convergence.Experimental results demonstrate that YOLOv8n-MAAI achieves a detection precision of 92.4%,a recall rate of 93.6%,and a mean average precision(mAP)of 94.3%,outperforming both the original YOLOv8n and other popular detection models such as YOLOv5s and YOLOv3-tiny.These improvements provide a more accurate and reliable solution for fire detection,supporting advancements in fire prevention and control technology.

曲娜;时尚;张晗;魏文龙

沈阳航空航天大学 安全工程学院,沈阳 110136沈阳航空航天大学 安全工程学院,沈阳 110136沈阳航空航天大学 安全工程学院,沈阳 110136沈阳航空航天大学 安全工程学院,沈阳 110136

资源环境

火灾图像检测目标检测YOLOv8多头自注意力机制损失函数

fire image detectiontarget detectionYOLOv8multi-head self-attention mechanismloss function

《沈阳航空航天大学学报》 2026 (2)

70-76,7

国家自然科学基金(项目编号:61901283)辽宁省自然科学基金(项目编号:2023-MS-241).

10.3969/j.issn.2095-1248.2026.02.010

评论