基于改进YOLOv7的小目标检测

doi:10.19678/j.issn.1000-3428.0065942

摘要/Abstract

摘要： 目前的目标检测技术已趋于成熟，但小目标检测仍是研究的难点。针对目标检测过程中小目标检测更容易出现漏检等问题，提出一种改进的YOLOv7目标检测模型。结合特征分离合并思想，对YOLOv7网络模型中的MPConv模块进行改进，以减少网络特征处理过程造成的特征损失，并通过实验确定放置改进MPConv模块的最佳位置。由于小目标检测过程中容易出现漏检的现象，利用ACmix注意力模块提高网络对小尺度目标的敏感度，降低噪声所带来的影响。在此基础上，使用SIoU替换原YOLOv7网络模型中的CIoU来优化损失函数，减少损失函数自由度，提高网络鲁棒性。在Okahublot公开的FloW-Img子数据集上进行实验，结果表明，对于数据集中的密集、小目标和超小目标三种情况的图片，改进后的YOLOv7网络模型相比原网络，漏检情况得到明显改善，且mAP达到71.1%，相比基线YOLOv7网络模型提升了4个百分点，检测效果优于原网络模型与传统经典目标检测网络模型。

关键词: 目标检测技术, 小目标检测, YOLOv7网络模型, 注意力模块, 损失函数

Abstract: Despite advancements in object detection technology, Small Object Detection(SOD) is still difficult to research.To address the challenge of easily missing detection in the process of object detection, this study proposes an improved YOLOv7 object detection model.Firstly, the MPConv module in the YOLOv7 model is improved by combining feature separation with merge, to reduce the feature loss caused by the process of network feature processing. The optimal position of the improved MPConv module is determined through experiments.Secondly, due to the phenomenon of missing detection in SOD, the algorithm uses the ACmix attention module to increase the sensitivity of the network to small-scale targets and reduce the influence caused by noise.Finally, SIoU is used to replace CIoU in the original YOLOv7 network model to optimize the loss function, reduce the freedom of the loss function, and improve the network robustness.Compared with the original network, the improved YOLOv7 network model can improve the missing detection situation of the images in the data set of the dense, small target, and ultra-small target by experimental comparison with the FloW-Img sub-dataset published by Okahublot.The results show that the mAP of the improved YOLOv7 network model can reach 71.1%, 4 percentage points higher than that of the baseline YOLOv7 network model, and the detection effect is better than that of the original network model and traditional classical target detection networks model.

Key words: object detection technology, Small Object Detection(SOD), YOLOv7 network model, attention module, loss function

中图分类号:

TP391

戚玲珑, 高建瓴. 基于改进YOLOv7的小目标检测[J]. 计算机工程, 2023, 49(1): 41-48.

QI Linglong, GAO Jianling. Small Object Detection Based on Improved YOLOv7[J]. Computer Engineering, 2023, 49(1): 41-48.

http://www.ecice06.com/CN/Y2023/V49/I1/41

图/表 11

20230701174918

20230701174922

20230701174926

20230701174929

20230701174932

20230701175012

20230701175016

20230701175019

20230701175023

20230701175026

20230701175029

参考文献

[1] 高新波, 莫梦竟成, 汪海涛, 等.小目标检测研究进展[J].数据采集与处理, 2021, 36(3):391-417. GAO X B, MO M J C, WANG H T, et al.Recent advances in small object detection[J].Journal of Data Acquisition and Processing, 2021, 36(3):391-417.(in Chinese)
[2] 谷永立, 宗欣欣.基于深度学习的目标检测研究综述[J].现代信息科技, 2022, 6(11):76-81. GU Y L, ZONG X X.A review of object detection study based on deep learning[J].Modern Information Technology, 2022, 6(11):76-81.(in Chinese)
[3] CHEN C, LIU M Y, TUZEL O, et al.R-CNN for small object detection[C]//Proceedings of IEEE International Conference on Computer Vision.Washington D.C., USA:IEEE Press, 2016:214-230.
[4] KRISHNA H, JAWAHAR C V.Improving small object detection[C]//Proceedings of the 4th IAPR Conference on Pattern Recognition.Washington D.C., USA:IEEE Press, 2017:340-345.
[5] ZHANG W, WANG S, THACHAN S, et al.Deconv R-CNN for small object detection on remote sensing images[C]//Proceedings of IEEE International Geoscience and Remote Sensing Symposium.Washington D.C., USA:IEEE Press, 2018:2483-2486.
[6] 赵加坤, 孙俊, 韩睿, 等.基于改进的Faster RCNN遥感图像目标检测[J].计算机应用与软件, 2022, 39(5):192-196, 290. ZHAO J K, SUN J, HAN R, et al.Object detection based on improved Faster RCNN for remote sensing image[J].Computer Applications and Software, 2022, 39(5):192-196, 290.(in Chinese)
[7] 贾可心, 马正华, 朱蓉, 等.注意力机制改进轻量SSD模型的海面小目标检测[J].中国图象图形学报, 2022, 27(4):1161-1175. JIA K X, MA Z H, ZHU R, et al.Attention-mechanism-based light single shot multiBox detector modelling improvement for small object detection on the sea surface[J].Journal of Image and Graphics, 2022, 27(4):1161-1175.(in Chinese)
[8] HAN J M, DING J, XUE N, et al.ReDet:a rotation-equivariant detector for aerial object detection[C]//Proceedings of IEEE/CVF Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2021:2785-2794.
[9] ZAND M, ETEMAD A, GREENSPAN M.Oriented bounding boxes for small and freely rotated objects[J].IEEE Transactions on Geoscience and Remote Sensing, 2022, 60(5):1-15.
[10] YU D H, XU Q, GUO H T, et al.Anchor-free arbitrary-oriented object detector using box boundary-aware vectors[J].IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2022, 15:2535-2545.
[11] ZHU X K, LÜ S C, WANG X, et al.TPH-YOLOv5:improved YOLOv5 based on transformer prediction head for object detection on drone-captured scenarios[C]//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision Workshops.Washington D.C., USA:IEEE Press, 2021:2778-2788.
[12] FU H X, SONG G Q, WANG Y C.Improved YOLOv4 marine target detection combined with CBAM[J].Symmetry, 2021, 13(4):623.
[13] BENJUMEA A, TEETI I, CUZZOLIN F, et al.YOLO-Z:improving small object detection in YOLOv5 for autonomous vehicles[EB/OL].[2022-09-01].https://arxiv.org/abs/2112.11798.
[14] WANG C Y, BOCHKOVSKIY A, LIAO H Y M.YOLOv7:trainable bag-of-freebies sets new state-of-the-art for real-time object detectors[EB/OL].[2022-09-01].https://arxiv.org/abs/2207.02696.
[15] YU J, ZHANG W.Face mask wearing detection algorithm based on improved YOLO-v4[J].Sensors, 2021, 21(9):3263.
[16] SONG Q, LI S, BAI Q, et al.Object detection method for grasping robot based on improved YOLOv5[J].Micromachines, 2021, 12(11):1273.
[17] NIU Z Y.A review on the attention mechanism of deep learning[J].Neurocomputing, 2021, 452:48-62.
[18] BRAUWERS G, FRASINCAR F.A general survey on attention mechanisms in deep learning[EB/OL].[2022-09-01].https://arxiv.org/abs/2203.14263.
[19] ZHOU D F, FANG J, SONG X B, et al.IoU loss for 2D/3D object detection[C]//Proceedings of 2019 International Conference on 3D Vision.Washington D.C., USA:IEEE Press, 2019:85-94.
[20] REZATOFIGHI H, TSOI N, GWAK J, et al.Generalized intersection over union:a metric and a loss for bounding box regression[C]//Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2019:658-666.
[21] ZHENG Z H, WANG P, LIU W, et al.Distance-IoU loss:faster and better learning for bounding box regression[J].Artificial Intelligence, 2020, 34(7):12993-13000.
[22] ZHENG Z H, WANG P, REN D W, et al.Enhancing geometric factors in model learning and inference for object detection and instance segmentation[J].IEEE Transactions on Cybernetics, 2022, 52(8):8574-8586.
[23] DAI Z H, LIU H X, LE Q V, et al.CoAtNet:marrying convolution and attention for all data sizes[EB/OL].[2022-09-01].https://arxiv.org/abs/2106.04803.
[24] PAN X R, GE C J, LU R, et al.On the integration of self-attention and convolution[C]//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Washington D.C., USA:IEEE Press, 2022:805-815.
[25] GEVORGYAN Z.SIoU loss:more powerful learning for bounding box regression[EB/OL].[2022-09-01].https://arxiv.org/abs/2205.12740.

选择文件类型/文献管理软件名称

选择包含的内容