针对无人机平台目标检测中目标尺寸大小不一、密集程度高、计算资源受限以及环境背景复杂等问题,基于YOLOv8框架提出轻量化与特征强化相结合的FCCD‑YOLO模型。首先,在主干网络中设计C2f‑FC模块,通过动态门控机制,有效降低模型参数与浮点运算量的同时保持关键空间特征提取能力;其次,构建DCGF(DySample Context‐Aware Guided Fusion)“三明治”结构,依次融合DySample、内容感知上下文融合（CGF）与C2f模块,增强多尺度特征交互与细节表达;最后,在检测层引入DS‑Head提升对微小目标的边界定位与类别判别能力。在VisDrone2019数据集上进行测试,实验结果表明,较基线模型YOLOv8s,FCCD‑YOLO实现了P提升1.16%、Recall提升1.90%、mAP50提升2.37%、mAP50:95提升1.68%,同时模型参数量与GFLOPS分别下降28.14%和20.35%,证明了该模型在资源受限的空中场景的实用性。
Abstract
Addressing the challenges of varying object sizes, high target density, limited onboard hardware, and complex backgrounds in UAV-based detection, this paper presents FCCD-YOLO—a lightweight, feature-enhanced model built on the YOLOv8 framework. First, we introduce a C2f FC module in the backbone network that uses a dynamic gating mechanism to substantially reduce model parameters and FLOPs while retaining essential spatial feature extraction capabilities. Next, we construct a "sandwich" DCGF (DySample Context-Aware Guided Fusion) structure that sequentially fuses DySample, Context-Guided Fusion (CGF), and the C2f module to enhance multi-scale feature interaction and fine-detail representation. Finally, we incorporate a DS Head into the detection layer to improve boundary localization and classification accuracy for minute targets. Experimental results on the VisDrone2019 dataset show that, compared with the baseline YOLOv8s, FCCD-YOLO achieves a 1.16% increase in Precision, a 1.90% increase in Recall, a 2.37% improvement in mAP50, and a 1.68% boost in mAP50:95, while reducing model parameters and GFLOPS by 28.14% and 20.35%, respectively. These gains demonstrate FCCD-YOLO’s effectiveness and suitability for resource-constrained aerial imaging scenarios.
关键词
YOLOv8 /
无人机 /
小目标检测 /
DCGF“三明治”结构 /
轻量化
Key words
YOLOv8 /
Unmanned Aerial Vehicles (UAVs) /
small object detection /
DCGF "sandwich" structure /
lightweight
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] Tan L,Liu Z,Huang X,et al.A transformer-based UAV instance segmentation model TF-YOLOv7[J].SIViP,2024(18):3299-3308.
[2] M Zheng,P Gao,R Zhang,et al.End-to-end object detection with adaptive clustering transformer[J].2020,arXiv:2011.09315.
[3] A Bochkovskiy,C Y Wang,H Y M Liao,YOLOv4:Optimal speed and accuracy of object detection[J].2020,arXiv:2004.10934.
[4] 李红岩,徐保庆,张子扬,等.基于全局上下文信息的遥感图像小目标检测[J].光学学报,2024,44(24):1-16.
[5] 陈海秀,陈子昂,宁馨,等.改进YOLOv10n的密集行人小目标检测方法[J/OL].小型微型计算机系统,1-10[2025-07-23].
[6] Li Y,Zhang X,Zhou Z.DBS-YOLO:a vehicle detection model based on improved YOLOv8 for UAV aerial scenes[C]//5th International Conference on Computer Vision,Image and Deep Learning (CVIDL),IEEE,2024:1432-1438.
[7] 郭越,杨江涛,刘志.复杂场景下的无人机小目标检测算法[J].工业控制计算机,2024,37(9):33-34+37.
[8] 田鹏,毛力.改进YOLOv8 的道路交通标志目标检测算法[J].计算机工程与应用,2024,60(8):202-212.
[9] 李和平,陈中举,许浩然,等.改进YOLO的钢材表面缺陷检测算法[J].现代电子技术,2024,47(13):7-14.
[10] R Varghese,M Sambath.YOLOv8:A Novel Object Detection Algorithm with Enhanced Performance and Robustness[C]//in 2024 International Conference on Advances in Data Engineering and Intelligent Computing Systems (ADICS),IEEE,2024,pp.1-6.
[11] Han Y,Wang C,Luo H,et al.LRDS-YOLO enhances small object detection in UAV aerial images with a lightweight and efficient design.Sci Rep.2025 Jul 2;15(1):22627.
[12] R.Varghese and S.M.YOLOv8:A Novel Object Detection Algorithm with Enhanced Performance and Robustness[C]//2024 International Conference on Advances in Data Engineering and Intelligent Computing Systems (ADICS),Chennai,India,2024,pp.1-6.
[13] Chen J,Kao S,He H,et al.Run,don't walk:chasing higher FLOPS for faster neural networks[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.2023:12021-12031.
[14] Shi D.Transnext:Robust foveal visual perception for vision transformers[C]//Proceedings of the IEEE/CVF conference on computer vision and pattern recognition.2024:17773-17783.
[15] L Yang,R Zhang,L Li,et al.SimAM:A simple, parameter-free attention module for convolutional neural networks[J].in Proc.Int.Conf.Mach.Learn.,Jul.2021,pp. 11863-11874.
[16] Hu J,Li S,Gang S.Squeeze-and-excitation networks[C]//Proceedings of the IEEE conference on computer vision and pattern recognition.2018.
[17] D Du,Y Zhang,L Bo.VisDrone-SOT2019:The vision meets drone single object tracking challenge results[C]//in Proc.IEEE/CVF Int.Conf.Comput.Vis.Workshops,Jul.2019,pp.213-226.
