Journal of Applied Optics, Volume. 45, Issue 4, 732(2024)
Small object detection algorithm based on improved YOLOv3
Figures & Tables(11)
Fig. 2. Diagram of Mosaic, Mixup and combination data enhancement example
Fig. 4. Schematic diagram of CSP module, SimSPPF module and feature enhancement module
Fig. 6. Comparison of detection effect of two algorithms on multi-target images
Table 1. Experimental environment configuration
View tableView in ArticleTable 1. Experimental environment configuration
配置 深度学习框架 操作系统 CPU GPU 运行内存 CUDA 型号 Pytorch 1.13 Windows 10 Inter Core i7-7770 Tesla P100 16 GB 11.3
Table 2. Comparison of detection effects by different algorithms
View tableView in ArticleTable 2. Comparison of detection effects by different algorithms
算法类别 主干网络 mAP/% FPS GPU SSD 300 VGG16 78.53 71 Tesla P100 YOLOv3 Darknet53 81.34 49 Tesla P100 文献[ 14 ]算法Darknet53 83.26 22 RTX 2070 文献[ 15 ]算法Darknet53 83.8 32 TITAN RTX 文献[ 16 ]算法VGG16 79.9 40 GTX 1080Ti 文献[ 17 ]算法Darknet53 84.6 19 RTX 2080Ti YOLOv4 CSPDarknet53 85.2 51 Tesla P100 本文算法 Darknet53 85.74 46 Tesla P100
Table 3. Comparison of missed detection rates for object detection
% View tableView in ArticleTable 3. Comparison of missed detection rates for object detection
% 类别 SSD YOLOv3 本文算法 sofa 22.18 23.8 18.4 bicycle 31 31.7 17.2 motorbike 21.5 24.3 16 dog 29 22.7 14.7 sheep 33 39.4 27.3 horse 16.1 17.5 11.5
Table 4. Effect comparison of different algorithms on different scale objects
% View tableView in ArticleTable 4. Effect comparison of different algorithms on different scale objects
% 类型 mAP R 小 (0,32] 中 (32,96] 大 (96,416] 小 (0,32] 中 (32,96] 大 (96,416] SSD 8.9 23.5 54 24.6 38.6 65.1 YOLOv3 10.7 25.3 45.3 26.1 42.1 56.3 YOLOv4 17.5 43 57.3 35.3 55.4 66.7 本文算法 19.4 35 59.2 41.6 48 68.1
Table 5. Results of ablation comparison experiment
View tableView in ArticleTable 5. Results of ablation comparison experiment
组别 算法改进方式 mAP/% FPS 小尺度 总体 1 YOLOv3 32.7 81.34 49 2 YOLOv3+数据增强 34.3 81.91 49 3 YOLOv3+数据增强+GIoU 37.1 83.02 48 4 YOLOv3+数据增强+GIoU+改进特征融合网络 42.5 84.64 46 5 YOLOv3+数据增强+GIoU+改进特征融合网络+特征增强模块 46.3 85.74 46
Tools
Get Citation
Copy Citation Text
Kai WANG, Shuli LOU, Yan WANG. Small object detection algorithm based on improved YOLOv3[J]. Journal of Applied Optics, 2024, 45(4): 732
Paper Information
Category: Research Articles
Received: May. 31, 2023
Accepted: --
Published Online: Oct. 21, 2024
The Author Email: LOU Shuli (娄树理)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20