Acta Photonica Sinica, Volume. 51, Issue 12, 1210003(2022)
Object Detection Method of Optical Remote Sensing Image with Multi-attention Mechanism
Figures & Tables(7)
Fig. 4. The loss value of MA-YOLOV5 network during training and validation
Table 1. Detection accuracy and speed of ARFCA with different branch numbers
View tableView in ArticleTable 1. Detection accuracy and speed of ARFCA with different branch numbers
Branch number of ARFCA Convolution kernel size
of each branch
mAP FPS 1 5×5 68.0 50.9 2 3×3，5×5 68.4 50.2 3 3×3，5×5，7×7 68.5 49.4 4 3×3，5×5，7×7，9×9 68.5 48.3
Table 2. The performance of different networks on the test set
View tableView in ArticleTable 2. The performance of different networks on the test set
Method Backbone mAP AP50 AP75 APS APM APL FPS SSD ResNet⁃50 48.4 80.9 51.5 29.0 57.5 51.3 77 RetinaNet ResNet⁃50 60.7 88.8 70.5 54.2 66.2 54.8 28 FCOS ResNet⁃50 63.5 89.8 74.3 56.2 69.3 58.2 38 YOLOv5 CSPdarknet 64.9 91.7 78.4 58.0 72.4 61.2 59 YOLOV5⁃STR CSPdarknet 66.3 92.6 80.2 58.6 73.3 62.9 52 YOLOv5⁃ARFCA CSPdarknet 67.2 93.0 80.6 59.2 73.5 63.4 54 MA⁃YOLOv5 CSPdarknet 68.5 93.4 82.8 60.3 76.5 65.3 49
Tools
Get Citation
Copy Citation Text
Hongjian FU, Hongyang BAI, Hongwei GUO, Yuman YUAN, Weiwei QIN. Object Detection Method of Optical Remote Sensing Image with Multi-attention Mechanism[J]. Acta Photonica Sinica, 2022, 51(12): 1210003
Paper Information
Category:
Received: May. 18, 2022
Accepted: Jul. 14, 2022
Published Online: Feb. 6, 2023
The Author Email: BAI Hongyang (hongyang@mail.njust.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20