改进YOLOv7的无边界不连续旋转检测算法

郭振江; 何小海; 吴晓红; 陈洪刚

doi:10.3969/j.issn.2096-6091.2023.20.001

您当前的位置：

首页 >

文章列表页 >

改进YOLOv7的无边界不连续旋转检测算法

研究论文 | 更新时间：2024-07-15

- 改进YOLOv7的无边界不连续旋转检测算法
- Enhanced Boundary-Free Discontinuous Rotation Detection Algorithm in YOLOv7
- 新一代信息技术 2023年6卷第20期页码：1-12
- 作者机构：
  
  四川大学电子信息学院，四川成都 610065
- 作者简介：
  
  [ "郭振江（1997—），男，现为四川大学硕士研究生。研究方向：图像处理，计算机视觉。E-mail：2895076272@qq.com" ]
  何小海（1964—），男，现为四川大学博士生导师。研究方向：图像处理与信息系统，模式识别，图像通信。E-mail：hxh@scu.edu.cn
- 基金信息：
  
  国家自然科学基金(62271336;62211530110);四川省科技计划资助(24QYCX0399);四川大学遂宁市校市战略合作“揭榜挂帅”科技项目(2022CDSN-15)
- DOI：10.3969/j.issn.2096-6091.2023.20.001
  中图分类号： TP391.41
- 纸质出版日期：2023-10-30，
- 稿件说明：
扫描看全文
郭振江, 何小海, 吴晓红, 等. 改进YOLOv7的无边界不连续旋转检测算法[J]. 新一代信息技术, 2023, 6(20): 01-12

GUO Zhen-jiang, HE Xiao-hai, WU Xiao-hong, et al. Enhanced Boundary-Free Discontinuous Rotation Detection Algorithm in YOLOv7[J]. New Generation of Information Technology, 2023, 6(20): 01-12
郭振江, 何小海, 吴晓红, 等. 改进YOLOv7的无边界不连续旋转检测算法[J]. 新一代信息技术, 2023, 6(20): 01-12 DOI： 10.3969/j.issn.2096-6091.2023.20.001.

GUO Zhen-jiang, HE Xiao-hai, WU Xiao-hong, et al. Enhanced Boundary-Free Discontinuous Rotation Detection Algorithm in YOLOv7[J]. New Generation of Information Technology, 2023, 6(20): 01-12 DOI： 10.3969/j.issn.2096-6091.2023.20.001.

摘要

针对遥感图像中目标方向、尺度和长宽比变化剧烈以及时而密集时而稀疏排列等造成的特征提取困难和基于回归的旋转检测器中存在的边界不连续问题，我们提出了以下解决方案：首先，使用YOLOv7检测器进行特征提炼与融合；其次，结合密集编码标签（Densely Coded Label，DCL），构建角度分类编码-解码器，实现旋转检测；然后，改进SimOTA自适应样本匹配策略为R-SimOTA，通过在代价函数中增加角度分类损失指导，提高样本分配准确性；最后，在损失函数中增加角度分类损失（DCL Loss）和权值（theta），并引入角度距离和长宽比敏感加权（Angle Distance and Aspect Ratio Sensitive Weighting，ADARSW），指导模型收敛并拟合出正确的角度预测值。在大型遥感图像数据集DOTA上进行了大量实验和视觉分析，结果表明该方案有效。

Abstract

Aiming at the difficulty of feature extraction caused by the drastic changes in object direction

scale and aspect ratio

and the sometimes dense and sometimes sparse arrangement in remote sensing images

and the boundary discontinuity problem in the regression-based rotating detector

we propose the following solutions: Firstly

the YOLOv7 detector is used for feature extraction and fusion. Secondly

in combination with densely coded label (DCL)

angle-classification codecs can be constructed to realize rotation detection. Then

the adaptive sample matching strategy of SimOTA is improved to R-SimOTA

and the accuracy of sample allocation is improved by adding the angle classification loss guidance to the cost function. Finally

angle distance and aspect ratio sensitive weighting (ADARSW) is introduced by adding angle classification loss (DCL Loss) and weight (theta) into the loss function. Guide the model to converge and fit the correct angle prediction value. A large number of experiments and visual analysis have been carried out on a large remote sensing image dataset DOTA

and the results show that the scheme is effective.

关键词

目标检测遥感图像旋转检测密集编码标签角度分类R-SimOTA自适应样本匹配

Keywords

object detectionremote sensing imagerotation detectiondense coding labelangle classificationR-SimOTA adaptive sample matching

references

REN S, HE K, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Trans Pattern Anal Mach Intell, 2017, 39(6): 1137-1149.

REDMON J, DIVVALA S, GIRSHICK R, et al. You only look once: Unified, real-time object detection[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2016: 779-788.

XIA G S, BAI X, DING J, et al. DOTA: A large-scale dataset for object detection in aerial images[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2018: 3974-3983.

LIN T Y, GOYAL P, GIRSHICK R, et al. Focal loss for dense object detection[C]//2017 IEEE International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2017: 2980-2988.

YANG X, YANG J R, YAN J C, et al. SCRDet: Towards more robust detection for small, cluttered and rotated objects[C]//2019 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2019: 8232-8241.

YANG X, YAN J. Arbitrary-oriented object detection with circular smooth label[C]//Proceedings of the 16th European Conference on Computer Vision—ECCV 2020, Part VIII 16. Glasgow: Springer, 2020: 677-694.

HE K M, GKIOXARI G, DOLLAR P, et al. Mask R-CNN[C]//2017 IEEE International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2017: 2961-2969.

GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//2014 IEEE Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2014: 580-587.

GIRSHICK R. Fast R-CNN[C]//2015 IEEE International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2015: 1440-1448.

LIN T Y, DOLLAR P, GIRSHICK R, et al. Feature pyramid networks for object detection[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2017: 2117-2125.

CHEN K, OUYANG W, LOY C C, et al. Hybrid task cascade for instance segmentation[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2019: 4974-4983.

HAN J M, DING J, XUE N, et al. ReDet: A rotation-equivariant detector for aerial object detection[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2021: 2786-2795.

SHI Q, ZHU Y, FANG C T, et al. RAOD: Refined oriented detector with augmented feature in remote sensing images object detection[J]. Applied Intelligence, 2022, 52(13): 15278-15294.

XIE X X, CHENG G, LI Q Y, et al. Fewer is more: Efficient object detection in large aerial images[J]. Science China Information Sciences, 2023, 67(1): 112106.

ZHENG S D, WU Z B, XU Y, et al. Learning orientation information from frequency-domain for oriented object detection in remote sensing images[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 1-12.

YANG X, YAN J, MING Q, et al. Rethinking rotated object detection with gaussian wasserstein distance loss[C]//International Conference on Machine Learning. San Diego: PMLR, 2021: 11830-11841.

YANG X, YANG X, YANG J, et al. Learning high-precision bounding box for rotated object detection via kullback-leibler divergence[J]. Advances in Neural Information Processing Systems, 2021, 34: 18381-18394.

YANG X, HOU L P, ZHOU Y, et al. Dense label encoding for boundary discontinuity free rotation detection[C]//2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2021: 15819-15829.

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[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2023: 7464-7475.

GE Z, LIU S, WANG F, et al. YOLOx: Exceeding yolo series in 2021[EB/OL]. (2021-08-06)[2024-02-08]. https://arxiv.org/ abs/2107.08430https://arxiv.org/abs/2107.08430

ZHENG Z H, WANG P, LIU W, et al. Distance-IoU loss: Faster and better learning for bounding box regression[J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2020, 34(7): 12993-13000.

ZHUANG W H, TANG X G, YANG G Y, et al. Remote sensing image object detection based on rotatable bounding box[C]//2021 International Conference on Digital Society and Intelligent Systems (DSInS). Piscataway: IEEE, 2021: 172-177.

LIU J, QIAO W S, XIONG Z L. OAB-YOLOv5: One-anchor-based YOLOv5 for rotated object detection in remote sensing images[J]. Journal of Sensors, 2022, 2022: 8515510.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

暂无数据