A novel transformer using dynamic range-enhanced discrete cosine transform for detecting bean leaf diseases.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Early detection of bean leaf diseases is vital for agricultural productivity.
• Subtle disease indicators can be challenging for human and machine detection.
• Effective disease identification supports crop management and food security.

Purpose of the Study:

• To develop an advanced machine learning approach for accurate bean leaf disease classification.
• To enhance feature extraction for improved disease detection in agriculture.
• To address limitations in current machine learning methods for plant pathology.

Main Methods:

• Proposed DCT-Transformers method combining dynamic range enhanced discrete cosine transform (DRE-DCT) preprocessing with Transformer models.
• DRE-DCT enhances image dynamic range by extracting high-frequency and subtle details.
• Transformer models classify bean leaf images pre- and post-DRE-DCT preprocessing.

Main Results:

• DCT-Transformers achieved 99.56% classification accuracy with preprocessed images, significantly higher than 95.92% with non-preprocessed images.
• The method outperformed existing state-of-the-art approaches (below 94%) and similar studies (below 98.5%).
• Enhanced feature extraction via DRE-DCT demonstrably improved disease classification performance.

Conclusions:

• The proposed DCT-Transformers method offers an efficient and highly accurate solution for early bean leaf disease detection.
• Improved feature extraction is key to advancing machine learning in agricultural disease diagnostics.
• This approach contributes to better disease management strategies and global food security.