改进YOLOX-m的火焰检测方法研究  

Research on Flame Detection Method Based on Improved YOLOX-m

作　　者：战乃岩[1] 张晓禾 姜泽旭 于儆芝 ZHAN Naiyan;ZHANG Xiaohe;JIANG Zexu;YU Jingzhi(School of Emergency Science and Engineering,Jilin Jianzhu University,Changchun 130119,China)

机构地区：[1]吉林建筑大学应急科学与工程学院,长春130119

出　　处：《计算机测量与控制》2025年第3期20-29,共10页Computer Measurement &Control

摘　　要：如今火灾自动报警技术已逐步朝着智能化、网络化和自动化的方向发展,然而,目前的火焰实时检测技术存在火焰实时检测精度低和网络计算参数量大等问题;针对以上问题,对YOLOX-m目标检测模型进行研究,提出改进YOLOX-m的轻量级火焰检测模型;通过将主干网络CSPDarknet-53替换为ShuffleNetV2,在降低计算量的同时提高网络精度,在ShuffleNetV2结构中插入RFB模块扩大感受野,在保持分辨率和精确定位检测目标的同时提升检测大目标的能力,将Neck部分的上采样替换为Pixel Shuffle以降低特征损失,为使网络能够关注到更关键的信息,增加注意力机制CBAM,从而提高模型整体性能;经过算法优化和实验测试,改进模型比YOLOX-m模型精度提高了2.87个百分点,参数量减少37.9%,计算量降低30.7%;改进模型成功应用于森林火灾、城市火灾等实际场景,通过对比可以相对更加精确地检测火焰。Nowadays,the automatic fire alarm technology is gradually developing towards intelligence,networking,and automation.However,current real-time flame detection techniques have the characteristics of low detection accuracy and high computational parameters.To address these issues,this paper focuses on the YOLOX-m object detection model,and proposes an improved lightweight fire detection model.The proposed model improves the YOLOX-m model by replacing the backbone network CSPDarknet-53 with the ShuffleNetV2,which reduces computational complexity while improving network accuracy.Additionally,the RFB module is inserted into the ShuffleNetV2 structure to increase the receptive field,enhancing the detection capability for large objects while maintaining the resolution and precise localization.The upsampling in the Neck is replaced with the Pixel Shuffle to minimize the feature loss.Furthermore,the attention mechanism CBAM is incorporated to enable the network to focus on the crucial information,thus improving the overall performance of the model.Through experimental testing of the optimized algorithm,the improved model increases the accuracy by 2.87%compared to the YOLOX-m model,with a reduction of 37.9%in parameters and a decrease of 30.7%in computational complexity.The improved model is successfully applied in real scenarios such as forest and urban fires,which can detect flames more accurately.

关 键 词：火焰检测 YOLOX 通道混洗 感受野增强 注意力机制 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]