航空兵器2025，Vol.32Issue(5)：83-91,9.DOI:10.12132/ISSN.1673-5048.2025.0070
GVCL-YOLO:融合Slim-Neck架构与CAA注意力的轻量化小目标检测模型
GVCL-YOLO:A Lightweight Small Object Detection Model Integrating Slim-Neck Architecture with CAA Attention
摘要
Abstract
In complex scenarios,small objects in images often suffer from insufficient feature representation due to limited scale and are highly susceptible to background noise interference.Existing YOLOv11n models still exhibit issues with missed and false detections when detecting small objects.To address these problems,this paper proposes an improved YOLOv11n model called GVCL-YOLO.Firstly,the GSConv and VoVGSCSP modules from the Slim-Neck architecture are employed for feature fusion.This effectively enhances the model's feature fusion capabilities and improves the quality of small object feature representation.Secondly,an additional small object detection layer is introduced to further strengthen the model's ability to extract low-level fine-grained features.Simultaneously,the CAA attention mechanism is integrated to suppress noise and enhance the robustness of small object detection.Furthermore,a lightweight LightDetectHead is designed to reduce the mo-del's computational complexity and improve inference efficiency.Finally,the Wise-MPDIoU loss function re-places the CIoU loss function to boost accuracy and accelerate convergence.Experimental results on the Vis-Drone2019 dataset demonstrate that GVCL-YOLO outperforms YOLOv11n,with mAP@0.5 and mAP@0.5:0.95 increasing by 4.2％and 2.7％respectively,while also reducing the number of parameters by 27.2％.关键词
目标检测/小目标/YOLOv11n/GVCL-YOLO/特征融合/轻量化Key words
object detection/small target/YOLOv11n/GVCL-YOLO/feature fusion/lightweight分类
武器工业引用本文复制引用
黄庆东,陈梓煌,苏宇辉,姚咏琪,刘依华..GVCL-YOLO:融合Slim-Neck架构与CAA注意力的轻量化小目标检测模型[J].航空兵器,2025,32(5):83-91,9.基金项目
国家自然科学基金项目(62201456) （62201456）
