[1]何宇豪,易明发,周先存,等.基于改进的Yolov5的无人机图像小目标检测[J].智能系统学报,2024,19(3):635-645.[doi:10.11992/tis.202210032]
　HE Yuhao,YI Mingfa,ZHOU Xiancun,et al.UAV image small-target detection based on improved Yolov5[J].CAAI Transactions on Intelligent Systems,2024,19(3):635-645.[doi:10.11992/tis.202210032]
点击复制

基于改进的Yolov5的无人机图像小目标检测

《智能系统学报》[ISSN 1673-4785/CN 23-1538/TP] 卷: 19 期数: 2024年第3期 页码: 635-645 栏目: 学术论文—机器人 出版日期: 2024-05-05
Title:
UAV image small-target detection based on improved Yolov5
作者:
何宇豪1, 易明发1, 周先存2, 王冠凌1
1. 安徽工程大学 电气工程学院, 安徽 芜湖 241000;
2. 皖西学院 电子与信息工程学院, 安徽 六安 237012
Author(s):
HE Yuhao1, YI Mingfa1, ZHOU Xiancun2, WANG Guanling1
1. College of Electrical Engineering, Anhui Polytechnic University, Wuhu 241000, China;
2. The Department of Electronic and Information Engineering, West Anhui University, Lu’an 237012, China
关键词:
图像处理; GhostConv卷积模块; 双向特征金字塔网络; 卷积块注意力模块; Soft双向特征金字塔网络; 轻量化模型; 小目标检测; VisDrone数据集
Keywords:
image processing; GhostConv convolution module; bidirectional feature pyramid network; convolutional block attention module; soft bidirectional feature pyramid network; lightweight model; small-target detection; VisDrone dataset
分类号:
TP391.4；V19
DOI:
10.11992/tis.202210032
文献标志码:
2023-12-04
摘要:
为了解决无人机航拍图像小目标检测算法检测速度与精度无法兼顾的问题，在Yolov5的基础上，提出了针对于无人机图像小目标检测的Yolov5_GBCS算法。在新的算法中，添加一个额外的检测头，以便增强对小目标的特征融合效果；在主干网络中分别采用GhostConv卷积模块、GhostBottleneckC3模块替换部分Conv模块和C3模块用以提取丰富特征和冗余特征以提高模型效率；引入加权双向特征金字塔网络（bidirectional feature pyramid network, BiFPN）结构，用以提高对小目标的检测精度；在主干网络和颈部网络中引入轻量化的卷积块注意力模块（convolutional block attention module, CBAM），关注重要特征并抑制不必要的特征，增强小目标特征表达能力；使用Soft-NMS算法来替换NMS，因此降低了小目标在密集场景下的漏检率。通过在VisDrone2019数据集上的实验结果表明，集成了所有改进的方法后的Yolov5_GBCS算法，不仅提高了检测精度，而且有效地提高了检测速度，模型的mAP从38.5%提高到43.2%，检测速度也从53 f/s提高到59 f/s。Yolov5_GBCS算法可以有效地实现无人机航拍图像中小目标识别。
Abstract:
The detection speed and accuracy of small targets captured by UAV aerial photography cannot be considered at the same time. To address this problem, a new algorithm based on the Yolov5 algorithm, namely Yolov5_GBCS, is proposed for small-target detection of UAV-captured images. In the new algorithm, an additional detector head is added to enhance the feature fusion effect of small targets. In the backbone network, the GhostConv convolution module and GhostBottleneckC3 module are used to replace some original Conv modules and C3 modules for extracting rich and redundant features, respectively, to improve the model efficiency. The weighted bidirectional feature pyramid network structure is introduced to enhance the detection accuracy of small targets. The lightweight convolutional block attention module is introduced into the backbone and neck networks to focus on important features and suppress unnecessary features to boost the ability of small-target feature expression. The Soft-NMS algorithm is used to replace the NMS for reducing the miss detection rate of small targets in dense scenes. Experimental results on the VisDrone 2019 dataset show that the Yolov5_GBCS algorithm integrating all improved methods enhances the detection accuracy and effectively improves the detection speed. The mAP of the subject model has been increased from 38.5% to 43.2%, and the detection speed from 53 f/s to 59 f/s. Therefore, the Yolov5_GBCS algorithm can effectively recognize small targets in the image captured by UAV aerial photography.
参考文献/References:
[1] 曹家乐, 李亚利, 孙汉卿, 等. 基于深度学习的视觉目标检测技术综述[J]. 中国图象图形学报, 2022, 27(6): 1697–1722
CAO Jiale, LI Yali, SUN Hanqing, et al. A survey on deep learning based visual object detection[J]. Journal of image and graphics, 2022, 27(6): 1697–1722
[2] LIU Chuanyang, WU Yiquan, LIU Jingjing, et al. MTI-YOLO: a light-weight and real-time deep neural network for insulator detection in complex aerial images[J]. Energies, 2021, 14(5): 1426.
[3] SAHIN O, OZER S. YOLODrone: improved YOLO architecture for object detection in drone images[C]//2021 44th International Conference on Telecommunications and Signal Processing. Brno: IEEE, 2021: 361–365.
[4] WANG Xuewen, ZHAO Qingzhan, JIANG Ping, et al. LDS-YOLO: A lightweight small object detection method for dead trees from shelter forest[J]. Computers and electronics in agriculture, 2022, 198: 107035.
[5] 彭继慎, 孙礼鑫, 王凯, 等. 基于模型压缩的ED-YOLO电力巡检无人机避障目标检测算法[J]. 仪器仪表学报, 2021, 42(10): 161–170
PENG Jishen, SUN Lixin, WANG Kai, et al. ED-YOLO power inspection UAV obstacle avoidance target detectionalgorithm based on model compression[J]. Chinese journal of scientific instrument, 2021, 42(10): 161–170
[6] 陈蕊, 郑华飞, 蒋鸿宇, 等. 结合仿真迁移学习和自适应融合的无人机小目标检测[J/OL]. 小型微型计算机系统, 2022: 1–10. (2022–05–10). https://kns.cnki.net/kcms/detail/21.1106.TP.20220509.1930.032.html.
CHEN Rui, ZHENG Huafei, JIANG Hongyu, et al. Combination of simulation-based transfer learning and adaptive fusion for UAV small object detection[J/OL]. Journal of Chinese computer systems, 2022: 1–10. (2022–05–10). https://kns.cnki.net/kcms/detail/21.1106.TP.20220509.1930.032.html.
[7] WOO S, PARK J, LEE J Y, et al. CBAM: convolutional block attention module[C]//European Conference on Computer Vision. Cham: Springer, 2018: 3–19.
[8] BODLA N, SINGH B, CHELLAPPA R, et al. Soft-NMS-improving object detection with one line of code[C]//2017 IEEE International Conference on Computer Vision. Venice: IEEE, 2017: 5562–5570.
[9] HU Jie, SHEN Li, SUN Gang. Squeeze-and-excitation networks[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 7132–7141.
[10] LIU Shu, QI Lu, QIN Haifang, et al. Path aggregation network for instance segmentation[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 8759–8768.
[11] KRISHNA K, NARASIMHA MURTY M. Genetic K-means algorithm[J]. IEEE transactions on systems, man, and cybernetics, part B (cybernetics), 1999, 29(3): 433–439.
[12] HE Kaiming, ZHANG Xiangyu, REN Shaoqing, et al. Deep residual learning for image recognition[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 770–778.
[13] DUAN Kaiwen, BAI Song, XIE Lingxi, et al. CenterNet: keypoint triplets for object detection[C]//2019 IEEE/CVF International Conference on Computer Vision. Seoul: IEEE, 2019: 6568–6577.
[14] WANG Haotong, LIU Bailin, LIU Zhiping, et al. LSS target threat estimation and capture prediction trajectory simulation[C]//2020 7th International Conference on Information Science and Control Engineering. Changsha: IEEE, 2020: 806–813.
[15] CAI Zhaowei, VASCONCELOS N. Cascade R-CNN: delving into high quality object detection[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018: 6154–6162.
[16] TIAN Zhi, SHEN Chunhua, CHEN Hao, et al. FCOS: fully convolutional one-stage object detection[C]//2019 IEEE/CVF International Conference on Computer Vision. Seoul: IEEE, 2019: 9626–9635.
[17] LI Yanghao, CHEN Yuntao, WANG Naiyan, et al. Scale-aware trident networks for object detection[C]//2019 IEEE/CVF International Conference on Computer Vision. Seoul: IEEE, 2019: 6053–6062.
[18] REN Shaoqing, HE Kaiming, GIRSHICK R, et al. Faster R-CNN: towards real-time object detection with region proposal networks[J]. IEEE transactions on pattern analysis and machine intelligence, 2017, 39(6): 1137–1149.
[19] ZHU Chenchen, HE Yihui, SAVVIDES M. Feature selective anchor-free module for single-shot object detection[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Long Beach: IEEE, 2019: 840–849.
[20] ZHANG Shifeng, CHI Cheng, YAO Yongqiang, et al. Bridging the gap between anchor-based and anchor-free detection via adaptive training sample selection[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Seattle: IEEE, 2020: 9756–9765.
[21] FENG Chengjian, ZHONG Yujie, GAO Yu, et al. TOOD: task-aligned one-stage object detection[C]//2021 IEEE/CVF International Conference on Computer Vision. Montreal: IEEE, 2021: 3490–3499.
[22] AHMED A, TANGRI P, PANDA A, et al. VFNet: a convolutional architecture for accent classification[C]//2019 IEEE 16th India Council International Conference. Rajkot: IEEE, 2019: 1–4.
相似文献/References:
[1]李庆武,蔡艳梅,徐立中.基于分块分类的智能视频监控背景更新算法[J].智能系统学报,2010,5(3):272.
　LI Qing-wu,CAI Yan-mei,XU Li-zhong.Background update algorithm based on blocks classification forintelligent video surveillance[J].CAAI Transactions on Intelligent Systems,2010,5():272.
[2]邵真天,袁杰.一种基于曲波变换的图像去块算法[J].智能系统学报,2012,7(2):102.
　SHAO Zhentian,YUAN Jie.An image deblocking algorithm based on curvelet transformation[J].CAAI Transactions on Intelligent Systems,2012,7():102.
[3]王锦榕,袁学海,刘增良.基于图像处理技术的瞳孔和角膜反射中心提取算法[J].智能系统学报,2012,7(5):423.
　WANG Jinrong,YUAN Xuehai,LIU Zengliang.An extraction method of pupil and corneal reflection centers based on image processing technology[J].CAAI Transactions on Intelligent Systems,2012,7():423.
[4]傅博,姜勇,王洪光,等.输电线路巡检图像智能诊断系统[J].智能系统学报,2016,11(1):70.[doi:10.11992/tis.201503043]
　FU Bo,JIANG Yong,WANG Hongguang,et al.Intelligent diagnosis system for patrol check images of power transmission lines[J].CAAI Transactions on Intelligent Systems,2016,11():70.[doi:10.11992/tis.201503043]
[5]任晓霞,孙秀明,耿鹏,等.多小波和NSDFB组合域递归滤波多聚焦图像融合[J].智能系统学报,2016,11(2):241.[doi:10.11992/tis.201509017]
　REN Xiaoxia,SUN Xiuming,GENG Peng,et al.Multifocus image fusion using a recursive filter in the combined domain of multiwavelets and NSDFB[J].CAAI Transactions on Intelligent Systems,2016,11():241.[doi:10.11992/tis.201509017]
[6]梁义辉,战强.一种面向无线图像传输的视觉平台[J].智能系统学报,2016,11(5):608.[doi:10.11992/tis.201512014]
　LIANG Yihui,ZHAN Qiang.A visual platform for wireless image transmission[J].CAAI Transactions on Intelligent Systems,2016,11():608.[doi:10.11992/tis.201512014]
[7]李霞丽,吴立成,樊艳明.易于硬件实现的压缩感知观测矩阵的研究与构造[J].智能系统学报,2017,12(3):279.[doi:10.11992/tis.201606037]
　LI Xiali,WU Licheng,FAN Yanming.Study and construction of a compressed sensing measurement matrix that is easy to implement in hardware[J].CAAI Transactions on Intelligent Systems,2017,12():279.[doi:10.11992/tis.201606037]
[8]姜婷,袭肖明,岳厚光.基于分布先验的半监督FCM的肺结节分类[J].智能系统学报,2017,12(5):729.[doi:10.11992/tis.201706018]
　JIANG Ting,XI Xiaoming,YUE Houguang.Classification of pulmonary nodules by semi-supervised FCM based on prior distribution[J].CAAI Transactions on Intelligent Systems,2017,12():729.[doi:10.11992/tis.201706018]
[9]王科俊,赵彦东,邢向磊.深度学习在无人驾驶汽车领域应用的研究进展[J].智能系统学报,2018,13(1):55.[doi:10.11992/tis.201609029]
　WANG Kejun,ZHAO Yandong,XING Xianglei.Deep learning in driverless vehicles[J].CAAI Transactions on Intelligent Systems,2018,13():55.[doi:10.11992/tis.201609029]
[10]刘彪,黄蓉蓉,林和,等.基于卷积神经网络的盲文音乐识别研究[J].智能系统学报,2019,14(1):186.[doi:10.11992/tis.201805002]
　LIU Biao,HUANG Rongrong,LIN He,et al.Research on braille music recognition based on convolutional neural networks[J].CAAI Transactions on Intelligent Systems,2019,14():186.[doi:10.11992/tis.201805002]

备注/Memo

收稿日期:2022-10-25。
基金项目:国家自然科学基金项目(61572366)；安徽高校自然科学研究重大项目(J2021ZD0116)；皖江高端装备制造协同创新中心开放基金项目(GCKJ2018013).
作者简介:何宇豪，硕士研究生，主要研究方向为目标检测。E-mail：1772307157@qq.com;易明发，硕士研究生，主要研究方向为目标检测。E-mail：1781987848@qq.com;王冠凌，教授，主要研究方向为嵌入式开发。主持省教育厅科研重点项目1项，获省科技进步奖一等奖1项，获发明专利授权4项，发表学术论文20余篇。E-mail：ahpu_405@163.com
通讯作者:王冠凌. E-mail：ahpu_405@163.com

更新日期/Last Update: 1900-01-01
Copyright © 《 智能系统学报》 编辑部
地址：（150001）黑龙江省哈尔滨市南岗区南通大街145-1号楼 电话：0451- 82534001、82518134　邮箱：tis@vip.sina.com