基于改进的YOLOv5遥感影像飞机目标检测

doi:10.13474/j.cnki.11-2246.2024.0813

摘要/Abstract

摘要： 针对现有目标检测算法在遥感影像飞机目标检测上存在检测速度慢、精度低、影像背景复杂、不易区分背景和目标,以及预测过程的收敛速度较慢、效率低等问题,本文在YOLOv5算法模型的基础上,采用引入注意力机制和更换损失函数两种优化策略,提升了算法对飞机目标的检测性能。在DOTA数据集上的训练结果显示,通过在YOLOv5网络结构主干部分的C3模块中引入CBAM注意力机制,算法检测性能得到显著提升。其中,模型训练结果的精确度(P)提升6%,召回率(R)提升2%,平均精度(mAP)提升2.8%;在YOLOv5预测过程中分别采用Focal EIoU和SIoU损失函数对原有CIoU损失函数进行替换,试验结果表明,改进后的算法模型回归精度显著提高,其中采用SIoU损失函数的模型优化效果最好,模型训练结果的精确度(P)提升4.3%,召回率(R)提升2%,平均精度值(mAP)提升2.7%。改进后的YOLOv5算法为实现对飞机目标的高精度实时检测提供参考。

关键词: 目标检测, 飞机检测, YOLOv5, 注意力机制, 损失函数

Abstract: In response to the problems of slow detection speed, low accuracy, complex image background, difficulty in distinguishing background and target, slow convergence speed and low efficiency in the prediction process of existing object detection algorithms in remote sensing image aircraft target detection, this paper adopts two optimization strategies, namely the introduction of attention mechanism and replacement loss function. Based on the YOLOv5 algorithm model, to improve the algorithm's detection performance for aircraft targets, the training results on the DOTA dataset show that by introducing the CBAM attention mechanism in the C3 module of the YOLOv5 network architecture backbone, the algorithm's detection performance has been significantly improved. Among them, the accuracy of the model training results P has been improved by 6%, the recall rate R has been improved by 2%, and the average precision(mAP) value has been improved by 2.8%. In the YOLOv5 prediction process, Focal EIoU and SIoU loss functions are used to replace the original CIoU loss functions. The experimental results show that the improved algorithm model significantly improve the regression accuracy, among which the model optimized using SIoU loss function have the best effect. The precision(P) of the model training results increased by 4.3%, the recall rate R increased by 2%, and the average precision(mAP) value increased by 2.7%. The improved YOLOv5 algorithm provides a reference for achieving high-precision real-time detection of aircraft targets.

Key words: object detection, aircraft inspection, YOLOv5, attention mechanism, loss function

中图分类号:

P237

黄子恒, 芮杰, 林雨准, 王淑香, 刘相云. 基于改进的YOLOv5遥感影像飞机目标检测[J]. 测绘通报, 2024, 0(8): 73-78,89.

HUANG Ziheng, RUI Jie, LIN Yuzhun, WANG Shuxiang, LIU Xiangyun. Aircraft target detection based on improved YOLOv5 in remote sensing imagery[J]. Bulletin of Surveying and Mapping, 2024, 0(8): 73-78,89.

参考文献

[1] 聂光涛,黄华.光学遥感图像目标检测算法综述[J].自动化学报,2021,47(8):1749-1768.
[2] 王威,李元垒,梁方,等.一种基于区域分割的地面飞机目标识别方法[J].计算机工程,2013,39(6):214-218.
[3] 朱丹,王斌,张立明.基于直线邻近平行性和GBVS显著性的遥感图像机场目标检测[J].红外与毫米波学报,2015,34(3):375-384.
[4] HUMMEL R A,KIMIA B,ZUCKER S W.Deblurring Gaussian blur[M]//Front-End Vision and Multi-Scale Image Analysis.[S.l.]:Springer Netherlands,2003:66-80.
[5] 高聪,王福龙.基于模板匹配和局部HOG特征的车牌识别算法[J].计算机系统应用,2017,26(1):122-128.
[6] 商立群,杜亚娟.Hu矩和Zernike矩在图象识别中的应用[J].西安科技学院学报,2000,20(1):53-56.
[7] 邹宁,柳健,周曼丽.基于Zernike矩的边缘检测方法[J].华中理工大学学报,1999,27(5):2-3.
[8] GIRSHICK R.Fast R-CNN[C]//Proceedings of 2015 IEEE International Conference on Computer Vision (ICCV).Santiago:IEEE,2015.
[9] GIRSHICK R,DONAHUE J,DARRELL T,et al.Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of 2014 IEEE Conference on Computer Vision and Pattern Recognition.[S.l.]:ACM Press,2014:580-587.
[10] HE Kaiming,ZHANG Xiangyu,REN Shaoqing,et al.Spatial pyramid pooling in deep convolutional networks for visual recognition[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2015,37(9):1904-1916.
[11] REDMON J,DIVVALA S,GIRSHICK R,et al.You only look once:unified,real-time object detection[C]//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).Las Vegas,NV:IEEE,2016:779-788.
[12] MAHTO P,GARG P,SETH P,et al.RefiningYolov4 for vehicle detection[J].International Journal of Advanced Research in Engineering and Technology(IJARET),2020,11(5):409- 419.
[13] HU Jie,SHEN Li,SUN Gang.Squeeze-and-excitation networks[C]//Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City:IEEE,2018.
[14] WANG Qilong,WU Banggu,ZHU Pengfei,et al.ECA-net:efficient channel attention for deep convolutional neural networks[C]//Proceedings of 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR).Seattle,WA:IEEE,2020.
[15] WOO S,PARK J,LEE J Y,et al.CBAM:convolutional block attention module[M]//Computer Vision-ECCV 2018.Cham:Springer International Publishing,2018:3-19.
[16] HOU Qibin,ZHOU Daquan,FENG Jiashi.Coordinate attention for efficient mobile network design[J].Proceedings of 2021 IEEE Conference on Computer Vision and Pattern Recognition,Nashville,TN;[s.n.],2021.
[17] ZHENG Zhaohui,WANG Ping,LIU Wei,et al.Distance-IoU loss:faster and better learning for bounding box regression[J].Computer Science,2019:208158250.
[18] ZHANG Yifan,REN Weiqiang,ZHANG Zhang,et al.Focal and efficient IOU loss for accurate bounding box regression[J].Neurocomputing,2022,506(C):146-157.
[19] GEVORGYAN Z.SIoU Loss:more powerful learning for bounding box regression[J].Computer Science,2022:249063031.
[20] XIA Guisong,BAI Xiang,DING Jian,et al.DOTA:a large-scale dataset for object detection in aerial images[C]//Proceedings of 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City,UT:IEEE,2018.