基于改进YOLOv8遥感目标检测算法研究  

Research on remote sensing target detection algorithm based on improved YOLOv8

作　　者：张宇阳 姜静 符珊 ZHANG Yuyang;JIANG Jing;FU Shan(Shenyang Ligong University,Shenyang 110159,China)

机构地区：[1]沈阳理工大学自动化与电气工程学院,辽宁沈阳110159

出　　处：《通信与信息技术》2025年第2期36-40,共5页Communication & Information Technology

摘　　要：针对遥感图像目标检测中复杂背景干扰、多尺度目标检测以及检测精度差的问题,提出一种基于改进YOLOv8的遥感图像目标检测算法CGF-YOLOv8。首先,将CPAM双重注意力模块嵌入特征提取网络之后,使模型提取的特征更充分,在处理复杂背景时更好地区分目标与非目标区域。其次,GFPN通过跨尺度特征融合的方式,将不同分辨率的特征信息有效地结合起来,使模型在处理多尺度目标时达到更高的检测准确性。最后,利用FIoU的锚框优化策略,通过赋予每个特征点不同的权重,在提升预测框与真实框匹配精度的同时,显著提高定位精度。在RSOD数据集上的测试结果表明,此方法的平均精度达到了98.6%,帧率FPS为每秒112帧。与原YOLOv8算法比,mAP提升了6.8%,帧率提升了12帧,达到了实时检测的效果。To address the challenges of complex background interference,multi-scale target detection,and poor detection accuracy in remote sensing image target detection,a remote sensing image target detection algorithm based on improved YOLOv8,named CGF-YO⁃LOv8,was proposed.Initially,the CPAM dual attention module is integrated after the feature extraction network,enabling the model to extract more comprehensive features and more effectively differentiate between target and non-target areas in complex backgrounds.Subsequently,the GFPN enhances detection accuracy for multi-scale targets by effectively combining feature information of different resolutions through cross-scale feature fusion.Finally,utilizing the FIoU anchor box optimization strategy,which assigns different weights to each feature point,not only improves the precision of matching predicted boxes with true boxes but also significantly enhances localization accuracy.Testing on the RSOD dataset demonstrated that this method achieved an average precision of 98.6%,an increase of 6.8%in mAP compared to the original YOLOv8 algorithm,with a frame rate of 250 frames per second(FPS),achieving real-time detection performance.

关 键 词：遥感图像 目标检测 注意力机制 全局特征融合 锚框优化 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]