基于YOLOv8的轻量级无人机航拍小目标检测模型  

A lightweight model for small object detection in UAV aerial images based on YOLOv8

作　　者：陈奕衡 廉佐政[1,2] CHEN Yi-heng;LIAN Zuo-zheng(College of Computer and Control Engineering,Qiqihar University,Heilongjiang Qiqihar 161006,China;Heilongjiang Provincial Key Laboratory of Big Data Network Security Detection and Analysis,Heilongjiang Qiqihar 161006,China)

机构地区：[1]齐齐哈尔大学计算机与控制工程学院,黑龙江齐齐哈尔161006 [2]黑龙江省重点实验室大数据网络安全检测分析,黑龙江齐齐哈尔市161006

出　　处：《齐齐哈尔大学学报(自然科学版)》2025年第3期37-42,共6页Journal of Qiqihar University(Natural Science Edition)

基　　金：黑龙江省省属高等学校基本科研业务费科研项目(145209126)。

摘　　要：针对无人机携带计算资源受限的问题,提出一种轻量化的YOLOv8小目标检测模型。在主干网络中引入AKConv,通过灵活的卷积核减少不必要的参数;构建新的颈部网络SBN,通过GSConv降低计算开销,并利用双向信息传递机制,提高对不同尺度目标的检测精度;使用Focal Modulation替代原有的SPPF模块,增强模型在图像关键区域的注意力机制。结果表明,改进后的模型较YOLOv8,参数量降低30%,mAP50提高2.3%,同时相较于目标检测领域常用的一些模型,在轻量化和检测精度上均有更好的表现,证明了改进的有效性。Aiming at the problem of limited computing resources carried by unmanned aerial vehicles,a lightweight small target detection model of YOLOv8 is proposed.AKConv is introduced into the backbone network to reduce unnecessary parameters through flexible convolution kernels.A new neck network SBN is constructed.The computational overhead is reduced through GSConv,and a bidirectional information transfer mechanism is used to improve the detection accuracy of targets of different scales.Focal Modulation is used to replace the original SPPF module to enhance the attention mechanism of the model in key areas of the image.Based on experiments,compared with YOLOv8,the improved model reduces the number of parameters by 30%and increases mAP50 by 2.3%.At the same time,compared with some commonly used models in the field of target detection,it has better performance in both lightweight and detection accuracy,proving the effectiveness of the improvement.

关 键 词：小目标检测 YOLOv8 AKConv SBN Focal Modulation 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]