针对目前城市背景烟火检测方法存在检测精度不高,易出现误检、漏检和耗时长等问题,提出一种基于YOLOv8s改进的烟火检测算法.引入加权双向特征金字塔(Bi-directional Feature Pyramid Network,BiFPN)增强特征融合,在BiFPN的基础上融合P2特征层提升小目标检测能力,同时添加基于跨空间学习的高效多尺度注意力(Efficient Multi-scale Attention,EMA)模块,突出目标特征同时抑制背景环境的干扰;为了有效利用特征图的语义信息,引入轻量级通用上采样算子——Content-Aware Re Assembly of Features(CARAFE);基于多尺度卷积注意力(Multi-scale Convolutional Attention,MCA)模块设计了一个轻量化的检测头并提升了检测精度;引入分组卷积空间金字塔池化SPPFCSPC_Group模块,在扩大感受野的同时具有更好的特征提取能力.实验结果表明,改进的YOLOv8s算法在基准模型的基础上计算量减少了 25％、参数量减少了37.6％、模型权重大小减少了 33.2％,平均精度均值(mean Average Precision,mAP)提升了 3.4％,基本满足烟火检测的需求.
Urban Background Smoke and Fire Detection Algorithm Based on Improved YOLOv8s
To solve the problems of low detection accuracy,prone to false detection,missing detection,and long processing time in current urban background smoke and fire methods,a smoke and fire detection algorithm based on improved YOLOv8s is proposed.The weighted Bi-directional Feature Pyramid Network(BiFPN)is introduced to enhance feature fusion.Building upon BiFPN,the P2 feature layer is fused to improve the detection ability for small targets.Additionally,an Efficient Multi-scale Attention(EMA)module based on cross-space learning is incorporated to highlight target features while suppressing background interference.To effectively utilize semantic information in feature maps,the lightweight universal upsampling operator Content-Aware ReAssembly of Features(CARAFE)is introduced.Furthermore,a lightweight detection head is designed based on the Multi-scale Convolutional Attention(MCA)module for improving detection accuracy.Finally,the SPPFCSPC_Group module,based on grouped convolution spatial pyramid pooling,is introduced to enlarge the receptive field and enhance feature extraction capability.Experimental results demonstrate that the improved YOLOv8s algorithm reduces computation by 25％,parameters by 37.6％,and model weight size by 33.2％compared to the baseline model,while increasing mean Average Precision(mAP)by 3.4％,thereby meeting the requirements for smoke and fire detection.
万方数据
维普
