CottonNet-MHA: a multi-head attention-based deep learning framework for cotton disease detection.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Cotton is a vital cash crop in India, threatened by diseases and pests.
• Accurate disease identification is crucial for maintaining crop health and yield.
• Current methods for disease detection can be time-consuming and less accurate.

Purpose of the Study:

• To develop a deep learning framework for automated cotton crop disease identification.
• To enhance disease detection accuracy and model interpretability.
• To create a practical tool for real-world agricultural applications.

Main Methods:

• Developed CottonNet-MHA, a novel deep learning framework utilizing multi-head attention mechanisms.
• Evaluated performance against benchmark models like VGG16, VGG19, InceptionV3, Xception, and MobileNet.
• Employed Gradient-weighted Class Activation Mapping (Grad-CAM) for model interpretability and visualization.
• Deployed the trained model via a web-based application for practical use.

Main Results:

• CottonNet-MHA demonstrated superior accuracy and efficiency in detecting cotton diseases compared to benchmark models.
• Multi-head attention mechanisms improved feature learning and highlighted diseased regions.
• Grad-CAM visualization confirmed the model's ability to focus on affected areas, enhancing trust.
• The web application facilitated real-world deployment and live disease monitoring.

Conclusions:

• CottonNet-MHA offers an automated, accurate, and interpretable solution for cotton disease diagnosis.
• The framework's success suggests potential for adaptation to other crop disease detection systems.
• The developed web platform supports practical implementation in agricultural settings, aiding farmers.