基于ECIOU结构嵌入YOLO的塔台视角目标检测  

Tower view object detection based on ECIOU structure embedded in YOLO

作　　者：钱基德 严浩 梁琰 曾昶畅 牟忆豪 QIAN Jide;YAN Hao;LIANG Yan;ZENG Changchang;MOU Yihao(Civil Aviation Flight Technology and Flight Safety Research Base,Civil Aviation Flight University of China,Guanghan 618307,China;Luoyang Beijiao Airport Co.,Ltd.,Luoyang 471099,China;CAAC Key Laboratory of Flight Technnology and Flight Safety,Guanghan 618307,China)

机构地区：[1]中国民用航空飞行学院民航飞行技术与飞行安全科研基地,广汉618307 [2]洛阳北郊机场有限责任公司,洛阳471099 [3]中国民用航空局飞行技术与飞行安全重点实验室,广汉618307

出　　处：《航空工程进展》2025年第2期70-78,共9页Advances in Aeronautical Science and Engineering

基　　金：民航飞行技术与飞行安全重点实验室项目(FZ2022ZX59,FZ2022KF03)。

摘　　要：现有的塔台视角目标检测系统易出现定位偏差大、小目标检测精度低等问题,为解决该问题,提出基于ECIOU结构嵌入YOLO v8模型的塔台视角下飞机类目标检测方法,以提高检测的准确性和效率。在传统YOLO v8模型基础上,增加CBAM模块,加强目标特征的判别性;引入GConv和SENet注意力机制,以优化Bottleneck结构,从而增强其特征提取能力;使用ECIOU Loss代替原有的CIOU损失函数,提升其在复杂环境下的检测性能;重新构建小目标检测头PWHead,以更好地捕捉小目标的细节信息。通过在Roboflow公开数据集上对模型进行评估,并将其性能与其他主流模型进行对比,结果表明:改进的YOLO v8模型的精确度达90.2%,平均精度均值达86.9%,较YOLO v8n分别提升了2.2%和1.3%,即提升了检测效率。In view of the problems that the existing tower view target detection system is prone to large positioning deviation and low small target detection accuracy,this paper proposes an aircraft target detection method based on the ECIOU structure embedded in the YOLO v8 model from the tower view to improve the accuracy and efficiency of detection.Based on the traditional YOLO v8 model,the CBAM module is first added to enhance the discrim-inability of target features.Then,the GConv and SENet attention mechanisms are introduced to optimize the Bot-tleneck structure to enhance its feature extraction ability.Thirdly,the ECIOU Loss is used to replace the original CIOU loss function to improve its detection performance in complex environments.Lastly,the small target detec-tion head PWHead is reconstructed to better capture the details of small targets.The model is evaluated on the Ro-boflow public dataset and its performance is compared with other mainstream models.The experimental results show that the accuracy of the improved YOLO v8 is 90.2%,and the average precision mean is 86.9%,which is 2.2%and 1.3%higher than that of YOLO v8n respectively,and the detection efficiency is improved.

关 键 词：YOLO v8 飞行安全 远程塔台 目标检测 图像处理 

分 类 号：V355[航空宇航科学与技术—人机与环境工程] V351.12