Ultra-lightweight tomatoes disease recognition method based on efficient attention mechanism in complex environment.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Existing tomato leaf disease recognition models struggle with real-world image complexity and single-leaf limitations.
• High hardware resource consumption hinders practical implementation in agriculture.

Purpose of the Study:

• To develop an integrated framework for tomato leaf detection and disease recognition.
• To create an ultra-lightweight model for efficient disease identification in diverse environments.

Main Methods:

• Developed an integrated framework combining leaf detection and disease recognition models.
• Engineered an ultra-lightweight recognition model using inverted residual modules and an efficient attention mechanism.
• Trained and validated the model on a dataset from real-world environments with 14 noise conditions.

Main Results:

• Achieved 97.84% accuracy with only 0.418 million parameters.
• Demonstrated enhanced recognition accuracy across 14 noisy environments compared to traditional models.
• Significantly reduced model parameters, overcoming limitations of single-disease image recognition.

Conclusions:

• The proposed framework effectively addresses limitations of existing tomato leaf disease recognition models.
• The ultra-lightweight model offers a practical and accurate solution for agricultural applications.
• Efficient attention mechanisms and optimized network architecture are key to balancing accuracy and resource consumption.