东南大学学报（英文版）2023，Vol.39Issue(4)：372-383,12.DOI:10.3969/j.issn.1003-7985.2023.04.006
基于YOLOX-Swin的高效建筑火灾烟雾检测和疏散模拟方法
Efficient fire-smoke detection and evacuation simulation from buildings based on YOLOX-Swin
摘要
Abstract
To achieve efficient emergency response and evacuation from building fires,the possibility of applying object detection technology to building fire emergency management is investigated.An application of object detection algorithms in the early warning stage of fire is proposed by combining a transformer,convolutional neural network,and lightweight attention mechanism module(namely convolutional block attention module)to extract the local and global features of flames and smoke,thereby improving the accuracy of the object detection algorithm and achieving fast localization of fire occurrence.An improved ant colony algorithm for path searching is proposed by improving the heuristic function and pheromone evaporation coefficient.A grid map model is developed in a case,and the effectiveness of the proposed method is verified through simulation and emulation by considering positioning information.The results show that compared with the YOLOX algorithm,the YOLOX-Swin model improves the average accuracy by 1.5％.The improved ant colony algorithm reduces the search range of the traditional ant colony algorithm,improves the convergence speed of the model,and avoids the problem of getting trapped in local optimum solutions.By integrating early warning of fire and personnel evacuation,a comprehensive building fire emergency management plan is developed.关键词
计算机视觉/自注意力/蚁群算法/火灾动力学模拟Key words
computer vision/self-attention/ant colony algorithm/fire dynamics simulator(FDS)分类
建筑与水利引用本文复制引用
徐照,戴天琦..基于YOLOX-Swin的高效建筑火灾烟雾检测和疏散模拟方法[J].东南大学学报（英文版）,2023,39(4):372-383,12.基金项目
The National Natural Science Foundation(No.72071043),the Natural Science Foundation of Jiangsu Province(BK20201280),Humanities and Social Science Fund of Ministry of Ed-ucation(20YJAZH114). （No.72071043）
