Super Resolution Generative Adversarial Network (SRGANs) for Wheat Stripe Rust Classification.

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Wheat yellow rust significantly impacts global crop yield and economic stability annually.
• Accurate and rapid disease detection is crucial, but often limited by low-resolution imagery.
• Existing detection models struggle with image quality constraints, hindering effective disease surveillance.

Purpose of the Study:

• To investigate the efficacy of super-resolution generative adversarial networks (SRGANs) for enhancing low-resolution wheat images.
• To improve the performance of deep learning models for wheat yellow rust detection using upsampled images.
• To demonstrate a practical solution for disease detection in resource-limited environments with poor image quality.

Main Methods:

• Image data preprocessing, including noise removal.
• Application of SRGANs to upscale low-resolution wheat images.
• Training convolutional neural networks (CNNs) on both original and SRGAN-upsampled images for wheat yellow rust detection.

Main Results:

• SRGANs effectively increased image resolution, aiding CNN feature learning.
• Models trained on SRGAN-enhanced images achieved 83% overall test accuracy.
• This represents a significant improvement over models trained on low-resolution images (75% accuracy).

Conclusions:

• SRGANs are a viable method for improving image quality in agricultural disease detection.
• The proposed approach enhances the performance of deep learning models for wheat yellow rust detection.
• This technique offers a valuable solution for real-world applications where high-resolution imaging is not feasible.