基于损失函数与注意力机制改进的YOLOv8火焰目标检测算法优化研究  

Optimization of YOLOv8 Flame Target Detection Algorithm Based on Improved Loss Function and Attention Mechanism

作　　者：周欣 徐培哲 李堃 熊椗宇 宋建平 夏子潮 ZHOU Xin;XU Pei-zhe;LI Kun;XIONG Ding-yu;SONG Jian-ping;XIA Zi-chao(Marine Design and Research Institute of China,Shanghai 200011,China;Faculty of Engineering,China University of Geoscience(Wuhan),Wuhan 430074,China;Power Engineering College,Naval University of Engineering,Wuhan 430033,China)

机构地区：[1]中国船舶及海洋工程设计研究院,上海200011 [2]中国地质大学(武汉)工程学院,武汉430074 [3]海军工程大学动力工程学院,武汉430033

出　　处：《船海工程》2025年第2期19-25,共7页Ship & Ocean Engineering

基　　金：海军工程大学自主研发计划(2023502060);中央高校基本科研业务费专项资金(162301222607,162301212668)。

摘　　要：针对火焰检测领域中YOLOv8模型精度不足的问题,提出两种改进方法以优化YOLOv8网络模型的火焰检测算法。设计一种改进的EIOU损失函数并引入YOLOv8模型中,通过对比实验确定参数α的最佳取值,使模型的收敛效果和火焰检测精度更佳,增强网络对不同场景下火焰的鲁棒性;引入AttnConv-EMA注意力机制,通过感知权重的非线性优化模型对内容的适应性,增强模型的精度和性能。使用自行建立的火焰检测数据集,基于Pytorch深度学习框架对YOLOv8模型进行训练,并结合不同的损失函数和注意力机制模块对原始的YOLOv8训练模型进行改进。研究结果表明,改进后的YOLOv8模型在火焰检测的检测精度上有显著提升,AttnConv-EMA注意力机制的引入进一步提升了模型的感知能力和精度,满足了火焰检测数据集的检测精度需求。Aiming at the lack of accuracy of YOLOv8 model in the field of flame detection,two improved methods were proposed to optimize the flame detection algorithm of YOLOv8 network model.An improved EIOU loss function was designed and introduced into the YOLOv8 model.The optimal value of parameterαwas determined through comparative experiments,which could improve the convergence effect and flame detection accuracy of the model,and enhance the robustness of the network to flame in different scenarios.The AttnConv-EMA attention mechanism was introduced to enhance the accuracy and performance of the model through the nonlinear optimization of the content adaptability of the model by sensing the weight.The self-established flame detection data set was used to train the YOLOv8 model based on the Pytorch deep learning framework,and the original YOLOv8 training model was improved by combining different loss function and attention mechanism modules.The research results showed that the improved YOLOv8 model has significantly improved the detection accuracy of flame detection,and the introduction of AttnConv-EMA attention mechanism further improves the perception ability and accuracy of the model,meeting the detection accuracy requirements of flame detection data sets.

关 键 词：YOLOv8 火焰检测 算法优化 

分 类 号：U662[交通运输工程—船舶及航道工程]