Tomato Leaf Disease Identification Framework FCMNet Based on Multimodal Fusion

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate tomato leaf disease recognition is vital for crop health and productivity.
• Single-mode disease identification methods lack accuracy and generalization.
• Multimodal fusion offers a promising approach to overcome these limitations.

Purpose of the Study:

• To propose a novel tomato leaf disease recognition framework, FCMNet, utilizing multimodal fusion.
• To enhance disease characteristic capture by integrating image and text data.
• To improve the accuracy and robustness of tomato disease identification.

Main Methods:

• Developed a Fourier-guided Attention Mechanism (FGAM) for enhanced feature expression and lesion localization.
• Introduced a Cross Vision-Language Alignment (CVLA) module for deep semantic interaction between image and text.
• Implemented a Multi-strategy Improved Coati Optimization Algorithm (MSCOA) for efficient training and parameter optimization.

Main Results:

• The FCMNet model demonstrated significant improvements in accuracy (2.61%), precision (2.85%), recall (3.03%), and F1 score (3.06%) compared to baseline models.
• FGAM enhanced feature stability and noise resistance through spectral transform.
• CVLA module improved multimodal semantic understanding.

Conclusions:

• FCMNet provides a superior solution for tomato leaf disease identification through effective multimodal fusion.
• The proposed methods offer enhanced accuracy, robustness, and generalization capabilities.
• This research has significant potential for practical agricultural applications and crop management.