重庆理工大学学报2026，Vol.40Issue(5)：84-90,7.DOI:10.3969/j.issn.1674-8425(z).2026.03.010
面向锥桶检测的改进YOLOv11n算法研究
Research on improved YOLOv11n algorithm for cone bucket detection
摘要
Abstract
This paper proposes an enhanced YOLOv11n algorithm to address the excessive parameters and low accuracy in cone detection for unmanned Formula Student racing cars.First,the neck network integrates a P2 small object detection layer to strengthen feature representation for distant small cones while the large object layer P5 is removed to reduce redundant computations for oversized objects.Then,the baseline model of the conventional convolutions used for downsampling are replaced by ADown modules,effectively compressing the model size without sacrificing detection accuracy.Next,the detection head is reconstructed using MBConv modules to design a lightweight MB_Detect head,further minimizing parameters and computational costs.Experimental results show,on the public FSACOCO dataset,the improved YOLOv11n algorithm increases the mean average precision from 90.4%to 94.8%,improves the recall rate from 83.5%to 90.1%,while reducing parameters and computational complexity to 1.35 Mand 5.1 GFLOPs respectively.Moreover,its memory footprint is down by 36%.Real-vehicle tests confirm the proposed method markedly reduces missed cone detections,providing a more robust environmental perception foundation for path planning in autonomous racing scenarios.关键词
深度学习/锥桶检测/下采样模块/轻量化检测头/小目标检测层Key words
deep learning/cone bucket detection/downsampling module/lightweight detection head/small object detection layer分类
交通工程引用本文复制引用
王希岳,李刚,刘杰,李旭,刘明伯,王铮..面向锥桶检测的改进YOLOv11n算法研究[J].重庆理工大学学报,2026,40(5):84-90,7.基金项目
辽宁省自然基金资助计划项目(2022-MS-376) （2022-MS-376）
辽宁省教育厅重点攻关项目(JYTZD2023081) （JYTZD2023081）
