基于改进YOLOv8n的红外行人车辆检测算法  

Infrared pedestrian vehicle detection algorithm based on improved YOLOV8

作　　者：秦海洋 谭功全 邓豪 王峣 蔡大洋 文力 QIN Hai-yang;TAN Gong-quan;DENG Hao;WANG Yao;CAI Da-yang;WEN Li(School of Automation and Information Engineering,Sichuan University of Science&Engineering,Yibin 644000,China;Artificial Intelligence Key Laboratory of Sichuan Province,Sichuan University of Science&Engineering,Yibin 644000,China)

机构地区：[1]四川轻化工大学自动化与信息工程学院,四川宜宾644000 [2]四川轻化工大学人工智能四川省重点实验室,四川宜宾644000

出　　处：《激光与红外》2025年第1期130-137,共8页Laser & Infrared

基　　金：人工智能四川省重点实验室科研项目(No.2019RYJ08)资助。

摘　　要：鉴于红外行人车辆图像分辨率低,质量不佳,噪声多等特点,检测难度较大,提出一种基于YOLOV8的红外图像行人车辆目标检测算法,即PSWG-YOLO。针对YOLOv8n网络,增加160×160的极大特征图P2提高模型对行人小目标的检测能力。同时,采用SPD-Conv部分代替原网络stride-2的卷积层,提升对低分辨率图像的特征提取能力。此外,将损失函数替换为WIoU,优化模型对低质量图像的处理。最后,引入Ghost模块降低模型复杂度。实验结果表明,改进后的PSWG-YOLO算法在保证较高的检测精度的同时,显著减少了模型体积和参数量。与原YOLOv8n算法在公开红外数据集FLIR_v2上P、R、mAP@0.5分别提升1.6%、6.3%、7.2%,且参数量减少16%,模型大小减少15.8%,提高了红外场景下行人车辆检测的精度并易于部署。Given that infrared pedestrian-vehicle images are difficult to detect due to their low resolution,poor quality,and high noise,an infrared image pedestrian and vehicle target detection algorithm based on YOLOV8 is proposed,namely PSWG-YOLO.For the YOLOv8n network,a 160×160 maximum feature map P2 is added to improve the model′s detection ability of pedestrian small targets.At the same time,the SPD-Conv part is used to replace the stride-2 convolutional layer of the original network to improve the feature extraction capability of low-resolution images.In addition,the loss function is replaced with WIoU to optimize the model′s processing of low-quality images.Finally,the Ghost module is introduced to reduce model complexity.The experimental results show that the improved PSWG-YOLO algorithm significantly reduces the model volume and parameter amount while ensuring high detection accuracy.Compared with the original YOLOv8n algorithm,the P,R,and mAP@0.5 on the public infrared data set FLIR_v2 are increased by 1.6%,6.3%,and 7.2%respectively,and the number of parameters is reduced by 16%,and the model size is reduced by 15.8%,which improves the accuracy of the pedestrian-vehicle detection in infrared scenarios and is easy to deploy.

关 键 词：红外目标检测 YOLOV8 SPD-Conv Ghost模块 损失函数 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TN219[自动化与计算机技术—计算机科学与技术]