摘要
当前矿井瓦斯-火灾早期预警通常依赖单一模态数据,存在片面性。因此,本文提出基于多模态融合与注意力机制的矿井瓦斯-火灾早期联合预警方法,采用可见光相机与热红外相机采集矿井环境多模态图像数据,并运用Retinex增强处理提升图像整体亮度;结合条带波变换算法与脉冲耦合神经网络,实现多模态图像融合,构建包含注意力机制的改进YOLOv8神经网络模型,实现矿井瓦斯-火灾早期联合预警。实验结果表明:该方法预警结果AUC值达到0.97,预警效果较好。
At present,mine gas-fire early warning usually relies on single-modal data,which is one-sided.Therefore,this paper proposes a joint early warning method for mine gas-fire based on multimodal fusion and an attention mechanism,which uses visible light cameras and thermal infrared cameras to collect multimodal image data of mine environments,and applies Retinex enhancement processing to improve the overall brightness of the images.The Bandelet transform algorithm combined with a pulse coupled neural network(PCNN)to realize multimodal image fusion,and an improved YOLOv8 neural network model with attention mechanism is established to achieve early joint warning of mine gas-fire.The experimental results show that the AUC value of the warning result of this method reaches 0.97,indicating a good warning effect.
作者
肖国亮
杨春虎
牛勇
杨博
Xiao Guoliang;Yang Chunhu;Niu Yong;Yang Bo(Huaneng Qingyang Coal and Electricity Co.,Ltd.Hetaoyu Coal Mine,Qingyang,Gansu 745000,China;Gansu Huating Coal and Electricity Co.,Ltd.)
出处
《计算机时代》
2026年第3期61-64,共4页
Computer Era
关键词
多模态融合
注意力机制
煤矿安全生产
火灾事故
联合预警
Multimodal Fusion
Attention Mechanism
Coal Mine Safety Production
Fire Accident
Joint Early Warning
