Please wait a minute...
浙江大学学报(工学版)
浙江大学学报(工学版)  2023, Vol. 57 Issue (6): 1224-1233    DOI: 10.3785/j.issn.1008-973X.2023.06.018
计算机与控制工程     
基于YOLOv5s的无人机密集小目标检测算法
韩俊(),袁小平*(),王准,陈烨
中国矿业大学 信息与控制工程学院，江苏 徐州 221116
UAV dense small target detection algorithm based on YOLOv5s
Jun HAN(),Xiao-ping YUAN*(),Zhun WANG,Ye CHEN
School of Information and Control Engineering, China University of Mining and Technology, Xuzhou 221116, China
 全文: PDF(2789 KB)   HTML
摘要：

针对无人机图像中背景复杂、小目标数量多且分布密集的特点，提出基于YOLOv5s的无人机密集小目标检测算法LSA_YOLO. 构造多尺度特征提取模块LM-fem，增强网络的特征提取能力. 为了抑制复杂背景的干扰，使算法关注目标信息，提出依靠多尺度上下文信息的、新的混合域注意力模块S-ECA. 设计自适应权重动态融合结构AFF，为浅层特征和深层特征合理分配融合权重. 将S-ECA、AFF应用于PANet结构，提高算法在复杂背景下的密集小目标检测能力. 使用损失函数Focal-EIOU代替损失函数CIOU，增强模型检测性能. 在公开数据集VisDrone2021上的实验结果表明，当设置输入分辨率为1 504 $ \times $1 504时，对所有目标类别的平均检测精度从YOLOv5s的51.5%提高到LSA_YOLO的57.6%.
关键词： 无人机;  小目标检测;  多尺度特征;  注意力机制;  特征融合    
Abstract:

The dense small target detection algorithm LSA_YOLO based on YOLOv5s for UAVs with complex backgrounds and multiples of small targets with dense distribution was proposed for UAV images. A multi-scale feature extraction module LM-fem was constructed to enhance the feature extraction capability of the network. A new hybrid domain attention module S-ECA relying on multi-scale contextual information has been put forward and a algorithm focus on target information was established aiming to suppress the interference of complex backgrounds. The adaptive weight dynamic fusion structure AFF was designed to assign reasonable fusion weights to both shallow and deep features. The capability of algorithm in detecting dense small targets in complex backgrounds was improved given the application of S-ECA and AFF in the structure of PANet. The loss function Focal-EIOU was utilized instead of the loss function CIOU to accelerate model detection efficiency. Experimental results on the public dataset VisDrone2021 public dataset show that the average detection accuracy for all target classes improves from 51.5% for YOLOv5s to 57.6% for LSA_YOLO when the set input resolution is set to 1 504 × 1 504.
Key words: UAV    small target detection    multi-scale features    attention mechanism    feature fusion
收稿日期: 2022-06-24 出版日期: 2023-06-30
CLC:  V 279  
基金资助: 国家科技支撑计划资助项目（2013BAK06B08）； 国家自然科学基金资助项目（32171241）
通讯作者: 袁小平     E-mail: m19816250697@163.com;1941@cumt.edu.cn
作者简介: 韩俊（1998—）男，硕士生，从事目标检测研究. orcid.org/0000-0001-8088-6777. E-mail： m19816250697@163.com
服务　　
把本文推荐给朋友
加入引用管理器
E-mail Alert
作者相关文章　　
韩俊
袁小平
王准
陈烨

引用本文:

韩俊,袁小平,王准,陈烨. 基于YOLOv5s的无人机密集小目标检测算法[J]. 浙江大学学报(工学版), 2023, 57(6): 1224-1233.

Jun HAN,Xiao-ping YUAN,Zhun WANG,Ye CHEN. UAV dense small target detection algorithm based on YOLOv5s. Journal of ZheJiang University (Engineering Science), 2023, 57(6): 1224-1233.

链接本文:

https://www.zjujournals.com/eng/CN/10.3785/j.issn.1008-973X.2023.06.018    或    https://www.zjujournals.com/eng/CN/Y2023/V57/I6/1224

图 1  LSA_YOLO算法的结构图
图 2  LM-fem和瓶颈层的结构对比图
图 3  挤压激励模块的结构图
图 4  空间位置注意力模块的结构图
图 5  混合域注意力模块的结构图
图 6  自适应权重动态融合模块的结构图
图 7  改进后的PANet结构图
图 8  VisDrone2021数据集中的数据信息
编号 模型 mAP $ {}_{50} $/% mAP $ {}_{50:95} $/% NP/106 O F/(帧·s?1) M/% E/%
1 YOLOv5s 33.2 16.7 6.978 15.5 125 57.5 47.2
2 YOLOv5s+LM-fem 33.9 17.5 6.920 14.9 130 55.0 45.3
3 YOLOv5s+CBAMbackbone 34.7 18.5 7.556 17.5 98 54.2 42.8
4 YOLOv5s+S-ECAbackbone 35.9 20.4 7.540 17.2 105 52.5 39.5
5 YOLOv5s+AFFneck 37.1 22.3 7.015 15.9 111 50.2 37.0
6 YOLOv5s+ $ \mathrm{S}\mathrm{A}\_\mathrm{P}\mathrm{A}\mathrm{N}\mathrm{e}\mathrm{t} $neck 37.8 22.9 8.135 18.4 95 48.9 35.2
7 YOLOv5s+Focal-EIOU[15] 33.5 17.2 7.322 16.1 120 56.6 45.8
8 LSA_YOLO 41.1 25.5 9.038 20.2 50 45.7 31.5
表 1  消融实验中各模型的检测性能评价指标
图 9  消融实验中各模型的平均精度均值
模型 mAP ${}_{50}/\text{%}$ NP/106 O F/(帧·s? 1)
$ \mathrm{Y}\mathrm{O}\mathrm{L}\mathrm{O}\mathrm{v}5\mathrm{s} $640 33.2 6.978 15.5 125
$ \mathrm{Y}\mathrm{O}\mathrm{L}\mathrm{O}\mathrm{v}5\mathrm{s} $1024 47.0 6.978 41.0 125
$ \mathrm{Y}\mathrm{O}\mathrm{L}\mathrm{O}\mathrm{v}5\mathrm{s} $1504 51.5 6.978 88.2 125
LSA_YOLO640 41.1 9.038 20.2 50
LSA_YOLO1024 49.7 9.038 51.7 50
LSA_YOLO1504 57.6 9.038 110.3 50
表 2  不同分辨率时模型的检测性能评价指标
算法 AP/% mAP $ {}_{50} $/%
A B C D E F G H I J
TridentNet[16] 22.8 9.0 5.3 46.2 30.7 25.5 21.3 16.0 39.0 17.9 43.1
RRNet[17] 30.5 14.8 14.1 51.5 35.8 35.2 28.8 19.0 45.0 26.0 55.0
CenterNet[18] 28.0 12.0 8.9 51.2 35.9 27.5 21.0 19.8 37.7 20.9 48.5
YOLOv5+head — — — — — — — — — — 33.8
YOLOv5+upsampling — — — — — — — — — — 50.5
YOLOv5+ M-Bi — — — — — — — — — — 43.6
YOLOv4[19] 25.0 13.1 8.5 64.2 22.5 22.6 11.5 8.0 44.5 22.0 43.0
YOLOv3-LITE[20] 34.6 22.9 8.0 71.2 31.4 22.1 15.5 7.1 41.3 32.7 41.9
MSC-CenterNet[21] 33.5 15.3 12.5 55.2 40.6 32.0 29.2 21.6 42.5 27.4 39.5
Faster R-CNN[22] 21.0 14.7 7.5 51.0 30.2 19.6 15.7 9.5 31.6 20.3 33.2
LSA_YOLO 37.2 25.4 18.5 58.6 35.7 35.8 29.4 21.5 47.2 28.4 57.6
表 3  不同算法在VisDrone2021数据集上的平均精度和平均精度均值
图 10  不同算法在VisDrone2021数据集上的平均精度均值柱状图
图 12  LSA_YOLO算法在复杂场景中的检测效果图
图 11  不同算法的目标检测效果可视化对比图
图 13  LSA_YOLO算法和基线算法的检测效果对比图
1 奉志强, 谢志军, 包正伟, 等 基于改进 YOLOv5的无人机实时密集小目标检测算法[J]. 航空学报, 2023, 44 (3): 327106
FENG Zhi-qiang, XIE Zhi-jun, BAO Zheng-wei, et al Real-time dense small object detection algorithm for UAV based on improved YOLOv5[J]. Acta Aeronautica et Astronautica Sinica, 2023, 44 (3): 327106
2 REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: unified, real-time object detection [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 779-788.
3 REDMON J, FARHADI A. YOLO9000: better, faster, stronger [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 6517-6525.
4 REDMON J, FARHADI A. YOLOv3: an incremental improvement [EB/OL]. [2022-06-20]. https://arxiv.org/abs/1804.02767.
5 BOCHKOVSKIY A, WANG C Y, LIA O H. YOLOv4: optimal speed and accuracy of object detection [EB/OL]. [2022-06-20]. https://arxiv.org/abs/2004.10934.
6 LIU W, ANGUELOV D, ERHAN D, et al. SSD: single shot multibox detector [C]// European Conference on Computer Vision. [S.l.]: Springer, 2016: 21-37.
7 李科岑, 王晓强, 林浩, 等 深度学习中的单阶段小目标检测方法综述[J]. 计算机科学与探索, 2022, 16 (1): 41- 58
LI Ke-cen, WANG Xiao-qiang, LIN Hao, et al Survey of one-stage small object detection methods in deep learning[J]. Journal of Frontiers of Computer Science and Technology, 2022, 16 (1): 41- 58
8 XIE L C, XUE Y L, YE J Z. UAV aerial photography target detection algorithm based on improved YOLOv5 [C]// Journal of Physics: Conference Series. [S.l.]: IOP Publishing, 2022, 2284(1): 012024.
9 YANG Y Z. Drone-view object detection based on the improved YOLOv5 [C]// Proceedings of the IEEE International Conference on Electrical Engineering, Big Data and Algorithms. Changchun: IEEE, 2022: 612-617.
10 吴萌萌, 张泽斌, 宋尧哲, 等. 基于自适应特征增强的小目标检测网络[J/OL]. 激光与光电子学进展, 2023, 60(6): 0610004. [20222-06-20]. https://www.opticsjournal.net/Articles/OJ7e6f90484b1776fd/References.
WU Meng-meng, ZHANG Ze-bin, SONG Yao-zhe, et al. Small object detection network based on adaptive feature enhancement [J/OL]. Advances in Laser and Opt-oelectronics, 2023, 60(66): 0610004. [2022-06-20]. https://www.opticsjournal.net/Articles/OJ7e6f90484b1776fd/References.
11 HU J, SHEN L, SUN G. Squeeze-and-excitation networks [C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 7132-7141.
12 LIN T Y, DOLLÁR P, GIRSHICK R, et al. Feature pyramid networks for object detection [C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu: IEEE, 2017: 2117-2125.
13 LI H, XIONG P, AN J, et al. Pyramid attention network for semantic segmentation [EB/OL]. [2022-06-20]. https://arxiv.org/abs/1805.10180.
14 LIU S, QI L, QIN H F, et al. Path aggregation network for instance segmentation [C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 8759–8768.
15 ZHANG Y F, REN W, ZHANG Z, et al Focal and efficient IOU loss for accurate bounding box regression[J]. Neurocomputing, 2022, 506: 146- 157
doi: 10.1016/j.neucom.2022.07.042
16 LI Y , CHEN Y , WANG N , et al. Scale-aware trident networks for object detection [C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Seoul: IEEE, 2019: 6054-6063.
17 CHEN C, ZHANG Y, LV Q, et al. RRNet: a hybrid detector for object detection in drone-captured images [C]// Proceedings of the IEEE/CVF International Conference on Computer Vision Workshop. Seoul: IEEE, 2019: 100-108.
18 ZHOU X, WANG D, KRAHENBUHL P. Objects as points [EB/OL]. [2022-06-20]. https://arxiv.org/abs/1904.07850.
19 ALI S, SIDDIQUE A, ATES H F, et al. Improved YOLOv4 for aerial object detection [C]// Proceedings of the 29th Signal Processing and Communications Applications Conference. Istanbul: IEEE, 2021: 1-4.
20 ZHAO H, ZHOU Y, ZHANG L, et al Mixed YOLOv3-LITE: a lightweight real-time object detection method[J]. Sensors, 2020, 20 (7): 1861
doi: 10.3390/s20071861
21 DU D W, WEN L Y, ZHU P F, et al. VisDrone-DET2020: the vision meets drone object detection in image challenge results [C]// Proceedings of the European Conference on Computer Vision. [S.l.]: Springer, 2020: 692-712.
[1] 项学泳,王力,宗文鹏,李广云. ASIS模块支持下融合注意力机制KNN的点云实例分割算法[J]. 浙江大学学报(工学版), 2023, 57(5): 875-882.
[2] 张剑钊,郭继昌,汪昱东. 基于融合逆透射率图的水下图像增强算法[J]. 浙江大学学报(工学版), 2023, 57(5): 921-929.
[3] 艾青林,崔景瑞,吕冰海,童桐. 基于损伤区域融合变换的轴承鼓形滚子表面损伤检测方法[J]. 浙江大学学报(工学版), 2023, 57(5): 1009-1020.
[4] 苏育挺,陆荣烜,张为. 基于注意力和自适应权重的车辆重识别算法[J]. 浙江大学学报(工学版), 2023, 57(4): 712-718.
[5] 卞佰成,陈田,吴入军,刘军. 基于改进YOLOv3的印刷电路板缺陷检测算法[J]. 浙江大学学报(工学版), 2023, 57(4): 735-743.
[6] 程艳芬,吴家俊,何凡. 基于关系门控图卷积网络的方面级情感分析[J]. 浙江大学学报(工学版), 2023, 57(3): 437-445.
[7] 曾耀,高法钦. 基于改进YOLOv5的电子元件表面缺陷检测算法[J]. 浙江大学学报(工学版), 2023, 57(3): 455-465.
[8] 杨帆,宁博,李怀清,周新,李冠宇. 基于语义增强特征融合的多模态图像检索模型[J]. 浙江大学学报(工学版), 2023, 57(2): 252-258.
[9] 刘超,孔兵,杜国王,周丽华,陈红梅,包崇明. 高阶互信息最大化与伪标签指导的深度聚类[J]. 浙江大学学报(工学版), 2023, 57(2): 299-309.
[10] 王林涛,毛齐. 基于RGB与深度信息融合的管片抓取位置测量方法[J]. 浙江大学学报(工学版), 2023, 57(1): 47-54.
[11] 凤丽洲,杨阳,王友卫,杨贵军. 基于Transformer和知识图谱的新闻推荐新方法[J]. 浙江大学学报(工学版), 2023, 57(1): 133-143.
[12] 刘近贞,陈飞,熊慧. 多尺度残差网络模型的开放式电阻抗成像算法[J]. 浙江大学学报(工学版), 2022, 56(9): 1789-1795.
[13] 王海军,张圣燕,杜玉杰. 响应和滤波器偏差感知约束的无人机目标跟踪算法[J]. 浙江大学学报(工学版), 2022, 56(9): 1824-1832.
[14] 郝琨,王阔,王贝贝. 基于改进Mobilenet-YOLOv3的轻量级水下生物检测算法[J]. 浙江大学学报(工学版), 2022, 56(8): 1622-1632.
[15] 莫仁鹏,司小胜,李天梅,朱旭. 基于多尺度特征与注意力机制的轴承寿命预测[J]. 浙江大学学报(工学版), 2022, 56(7): 1447-1456.