Acta Optica Sinica, Volume. 41, Issue 22, 2228001(2021)
Remote Sensing Image Caption Method Based on Attention and Reinforcement Learning
Figures & Tables(8)
Fig. 2. Visualization of the attention mechanism. (a) Original image; (b) attention image
Fig. 4. Some images and annotation captions in RSICD dataset. (a) Image; (b) annotation captions
Fig. 5. Remote sensing image caption results of different methods. (a) NIC; (b) Attend; (c) Adaptive; (d) Ours
Fig. 6. Visualization results of attention mechanism in remote sensing image caption. (a) Image 1; (b) image 2
Table 1. Image caption detection results of different methods
View tableTable 1. Image caption detection results of different methods
Method BLUE-1 BLUE-2 BLUE-3 BLUE-4 METEOR ROUGE_L CIDEr NIC[ 15 ]0.658 0.519 0.436 0.357 0.322 0.605 1.926 Attend[ 16 ]0.676 0.533 0.442 0.364 0.330 0.617 1.965 Adaptive[ 17 ]0.685 0.546 0.451 0.372 0.338 0.629 1.984 Ours 0.703 0.562 0.467 0.385 0.354 0.642 2.051
Table 2. Influence of different modules on remote sensing image caption
View tableTable 2. Influence of different modules on remote sensing image caption
Method BLUE-1 BLUE-4 METEOR ROUGE_L CIDEr Encode-Decode 0.667 0.361 0.325 0.612 1.958 Encode-Decode+Atten 0.681 0.369 0.336 0.625 1.977 Encode-Decode+Atten+Reinforce 0.703 0.385 0.354 0.642 2.051
Tools
Get Citation
Copy Citation Text
Yuanjun Nong, Junjie Wang. Remote Sensing Image Caption Method Based on Attention and Reinforcement Learning[J]. Acta Optica Sinica, 2021, 41(22): 2228001
Paper Information
Category: Remote Sensing and Sensors
Received: Apr. 30, 2021
Accepted: Jun. 3, 2021
Published Online: Nov. 17, 2021
The Author Email: Wang Junjie (wjj@ouc.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20