Automatic cassava disease recognition using object segmentation and progressive learning.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Cassava cultivation is crucial globally but faces significant threats from plant diseases.
• Existing cassava disease detection methods are often impractical for large-scale agricultural applications due to cost, time, and environmental limitations.

Purpose of the Study:

• To develop an efficient and accurate deep learning framework for early detection of cassava diseases.
• To enable real-world application of disease detection in diverse agricultural settings.

Main Methods:

• A novel self-supervised object segmentation technique was employed.
• A progressive learning algorithm (PLA) incorporating triplet loss and classification loss was utilized for robust feature embedding.
• The framework was evaluated on the Cassava Leaf Disease Classification (CLDC) dataset.

Main Results:

• The proposed deep learning framework achieved a high accuracy of 91.43% on the CLDC dataset.
• The method outperformed all other participants in the Kaggle competition for cassava leaf disease classification.
• Demonstrated superior performance in identifying cassava diseases under real-world conditions.

Conclusions:

• The developed deep learning framework provides a practical and efficient solution for cassava disease detection.
• The study highlights the potential for AI-driven tools in safeguarding global cassava production.
• This approach facilitates large-scale, real-world application in agriculture for improved crop management.