[1]舒 军,吴 柯.基于改进YOLOv3的航拍目标实时检测方法[J].湖北工业大学学报,2020,(1):21-24.
　SHU Jun,WU Ke.Research on Real Time Detection Method of Aerial Targets Based on Improved YOLOv3[J].,2020,(1):21-24.
点击复制

基于改进YOLOv3的航拍目标实时检测方法()
分享到：

《湖北工业大学学报》[ISSN:1003-4684/CN:42-1752/Z]

卷:
期数:
2020年第1期
页码:
21-24
栏目:
湖北工业大学学报
出版日期:
2020-02-28

文章信息/Info

Title:
Research on Real Time Detection Method of Aerial Targets Based on Improved YOLOv3
文章编号:
1003-4684(2020)01-0021-04
作者:
舒  军;  吴  柯
湖北工业大学电气与电子工程学院， 湖北 武汉 430068
Author(s):
SHU Jun; WU Ke
School of Electrical and Electronic Engineering, Hubei Univ. of Tech., Wuhan 430070, China
关键词:
航拍图像;  YOLOv3;  DOTAv1.0;  Kmeans++
Keywords:
aerial images;  YOLOv3;  DOTA v1.0;  K means++
分类号:
TP751.1
文献标志码:
A
摘要:
对于航拍图像中的小型目标，YOLOv3算法模型对其识别精准度低，在目标被遮挡或目标较密集时存在漏检现象。针对上述问题提出了一种基于改进YOLOv3的航拍目标实时检测方法，该方法加入104×104特征分辨率的检测模块并删减了13×13特征分辨率的检测模块，同时增加了浅层网络的层数，用于提取更加细微的像素特征；在训练阶段针对DOTA-v1.0航拍数据集使用K-means++聚类得到9个先验框进行检测，用于提升整体网络的训练速度。实验结果表明：改进后的YOLOv3检测算法的检出率提升了15.0%，mAP-50提升了10.5%。
Abstract:
For small targets in aerial images, the YOLOv3 algorithm model had low recognition accuracy, and there was a missed detection when the target was occluded or dense. Aiming at these problems, this paper proposed a real time detection method based on improved YOLOv3 for aerial targets. This method added a detection module of 104×104 feature resolution and cut the detection module of 13×13 feature resolution, and added shallow layers, which was used to extract more subtle pixel features. In the training phase, K means++ clustering was used to obtain 9 prior boxes for the DOTA v1.0 aerial dataset, which was used to improve the training speed of the overall network. Experiments showed that the detection rate of the improved YOLOv3 detection algorithm increased by 15.0%, and the mAP 50 increased by 10.5%.

参考文献/References:

[1] 房磊. 4K超高清数字发展动态研究[J].广播与电视技术, 2014, 41(12):1-2.
[2] Liu W, Anguelov D, Erhan D, et al. Ssd: Single shot multibox detector[C]//European conference on computer vision. Springer, Cham, 2016: 21-37.
[3] 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. 2016: 779-788.
[4] Wu B, Iandola F, Jin P H, et al. Squeezedet: Unified, small, low power fully convolutional neural networks for real-time object detection for autonomous driving[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. 2017: 129-137.
[5] Redmon J , Farhadi A . YOLOv3: An Incremental Improvement[J]. IEEE Conference on Computer Vision and Vision and Pattern Recognition. IEEE, 2018:89-95.
[6] Szegedy C, Liu W, Jia Y, et al. Going deeper with convolutions[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 1-9.
[7] Redmon J, Farhadi A. YOLO9000: better, faster, stronger[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 7263-7271.
[8] Simonyan, Karen, Zisserman, Andrew. Very Deep Convolutional Networks for Large-Scale Image Recognition[J]. Computer Science, 2014: 542-564.
[9] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770-778.
[10] 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. 2017: 2117-2125.
[11] Xia G S, Bai X, Ding J, et al. DOTA: A large-scale dataset for object detection in aerial images[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018: 3974-3983.
[12] Arthur D, Vassilvitskii S. k-means++: The advantages of careful seeding[C]//Proceedings of the eighteenth annual ACM-SIAM symposium on Discrete algorithms. Society for Industrial and Applied Mathematics, 2007: 1027-1035.

备注/Memo

备注/Memo:
［收稿日期］ 2019-09-29
［第一作者］ 舒   军(1973-)， 男， 湖北荆门人，工学博士，湖北工业大学副教授，研究方向为机器视觉，深度学习
更新日期/Last Update: 2020-04-11