基于改进YOLOv7的火焰烟雾识别  

Flame and Smoke Recognition Based on Improved YOLOv7

作　　者：雷景生 李顶 俞云祥 杨胜英 LEI Jing-sheng;LI Ding;YU Yun-xiang;YANG Sheng-ying(Zhejiang University of Science and Technology,Hangzhou Zhejiang 310023,China;Zhejiang Dingli Industrial Co.,LTD,Lishui Zhejiang 321400,China)

机构地区：[1]浙江科技学院,浙江杭州310023 [2]浙江鼎立实业有限公司,浙江丽水321400

出　　处：《计算机仿真》2025年第2期198-203,共6页Computer Simulation

基　　金：国家自然科学基金(61972357);新疆维吾尔自治区自然科学基金(2022D01C349)。

摘　　要：火焰和烟雾检测是一项重要的车间安全任务。然而,在复杂场景下同时进行火焰和烟雾检测的算法研究相对较少,且在实际应用中检测效果待提高。因此,提出了一种改进YOLOv7算法。首先,设计了特征提取SCF模块,它能够增强不同层特征图的表示能力。其次,模型融合了CBAM注意力机制,提高对不同尺度特征的关注度。最后,模型引人Focal Loss损失函数,优化预测框回归精度和网络鲁棒性。模型在含9462张图片的公开数据集上开展一系列对比实验。结果表明,改进后的模型在复杂场景下对火焰烟雾检测表现出色,在mAP0.5上,改进后的模型表现达到了48.8%,在mAP0.75上表现为17.8%,相比较YOLOv7原模型,精度上分别提升了5%和17.9%。Flame and smoke detection is an important workshop safety task.However,there are relatively few researches on algorithms for simultaneous flame and smoke detection in complex scenes,and the detection effect needs to be improved in practical applications.Therefore,an improved YOLOv7 algorithm is proposed in this paper.Firstly,a feature extraction SCF module is designed,which can enhance the representation ability of feature maps of different layers.Secondly,the model integrates the CBAM attention mechanism to improve the attention of features at different scales.Finally,the model introduced Focal Loss function to optimize the prediction frame regression accuracy and network robustness.The model conducted a series of comparative experiments on a publicly available data set of 9,462 images.The results show that the improved model performs well in flame and smoke detection in complex scenes.The performance of the improved model reaches 48.8%on mAP0.5 and 17.8%on mAP0.75.Compared with the 0-riginal model of YOLOv7,the accuracy of the improved model is improved by 5%and 17.9%respectively.

关 键 词：火焰烟雾检测 高效聚合网络 特征提取 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]