Related Experiment Video
Updated: Oct 27, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
586
Remote Sensing Image Dataset Expansion Based on Generative Adversarial Networks with Modified Shuffle Attention
Lu Chen1, Hongjun Wang1, Xianghao Meng1
1School of Electronic Countermeasures, National University of Defense Technology, Hefei 230000, China.
Sensors (Basel, Switzerland)
|July 24, 2021
Summary
This study introduces a modified Shuffle Attention Generative Adversarial Network (GAN) for expanding limited remote-sensing image datasets. The enhanced GAN generates higher quality, detailed aircraft images, addressing limitations of traditional methods.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Remote Sensing
Background:
- Neural networks are crucial for remote-sensing image processing but require large datasets.
- Expanding limited datasets is a significant research challenge.
- Generative Adversarial Networks (GANs) offer potential for data expansion.
Purpose of the Study:
- To develop a novel GAN model for generating high-quality remote-sensing images from limited data.
- To improve image quality and detail in generated samples.
- To address image distortion issues in GANs.
Main Methods:
- Modification of a Shuffle Attention network and its integration into a GAN framework.
- Introduction of an "equal stretch resize" method to prevent image distortion.
- Incorporation of a coordinate attention (CA) module for comparative analysis.
Main Results:
- The modified Shuffle Attention GAN generates more refined and high-quality diversified aircraft pictures.
- The model effectively produces detailed features even with limited input datasets.
- Qualitative and quantitative evaluations demonstrate superior performance compared to existing GANs.
Conclusions:
- The proposed Shuffle Attention GAN is effective for data expansion in remote-sensing.
- The equal stretch resize method successfully mitigates image distortion.
- This approach enables the generation of detailed and diverse imagery from constrained datasets.