A novel framework GRCornShot for corn disease detection using few shot learning with prototypical network.

Area of Science:

• Agricultural Science
• Computer Science
• Machine Learning

Background:

• Accurate and timely plant disease detection is crucial for global food security.
• Deep learning models for plant disease diagnosis require extensive labeled data, posing a challenge for real-world applications.

Purpose of the Study:

• To develop a novel model, GRCornShot, for corn disease diagnosis using few-shot learning.
• To address the data scarcity issue in training deep learning models for plant disease classification.

Main Methods:

• Proposed GRCornShot model utilizing Prototypical Networks and metric learning.
• Incorporated Gabor filter with ResNet-50 backbone for enhanced texture feature extraction.
• Employed few-shot learning strategies (4-way N-shot) for model training and evaluation.

Main Results:

• GRCornShot achieved high classification accuracies: 96.19% (2-shot), 96.54% (3-shot), 96.90% (4-shot), and 97.89% (5-shot).
• Demonstrated effective corn disease classification with significantly reduced labeled data requirements.
• Highlighted the robustness of Gabor filters in texture feature extraction for improved performance.

Conclusions:

• Few-shot learning offers a promising solution for plant disease detection in agronomic applications.
• GRCornShot provides a precise and data-efficient method for diagnosing corn diseases.
• The model's performance indicates its potential for practical implementation in precision agriculture.