Related Experiment Video
Updated: Jun 28, 2025
11:38
Author Spotlight: Enhancing PSC-to-Functional Cell Differentiation Using ML Models Based on Live-Cell Bright-Field Imaging
Published on: October 4, 2024
542
SM-CycleGAN: crop image data enhancement method based on self-attention mechanism CycleGAN
Dian Liu1, Yang Cao2, Jing Yang1,3
1School of Mechanical Engineering, Guizhou University, Guiyang, 550025, China.
Scientific Reports
|April 23, 2024
Summary
This study enhances crop image datasets using a self-attention generative adversarial network (GAN) to improve disease detection and growth stage analysis. The novel method boosts image quality, aiding agricultural AI applications.
Area of Science:
- Agricultural Science
- Computer Vision
- Artificial Intelligence
Background:
- Accurate crop disease detection and growth stage identification rely on extensive image data.
- Existing datasets suffer from asymmetry and poor image quality due to environmental factors, hindering model performance.
Purpose of the Study:
- To enhance crop image datasets for improved accuracy in disease detection and growth stage analysis.
- To address limitations of existing datasets, including asymmetry and image quality issues.
Main Methods:
- Proposed an innovative crop image data-enhancement method using recurrent generative adversarial networks (GANs) fused with a self-attention mechanism.
- Introduced a self-attention module into CycleGAN (SM-CycleGAN) to improve data correlation capture.
- Developed a new enhanced loss function to optimize model performance.
Main Results:
- SM-CycleGAN demonstrated improved perception and information capture capabilities compared to standard CycleGAN.
- Peak signal-to-noise ratio (PSNR) for tobacco and tea leaf images improved by 2.13% and 3.55%, respectively.
- Structural similarity index measure (SSIM) improved by 1.16% and 2.48% for tobacco and tea leaf images, respectively.
Conclusions:
- The proposed SM-CycleGAN effectively enhances crop image data quality and stability.
- This method offers a robust solution for improving agricultural AI tasks reliant on image analysis.
- The enhanced datasets contribute to more accurate crop monitoring and management.
