High-Performance Grape Disease Detection Method Using Multimodal Data and Parallel Activation Functions.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Grape disease detection is crucial for crop yield and quality.
• Existing methods often lack accuracy, speed, or robustness for field applications.

Purpose of the Study:

• To develop and evaluate a novel deep learning model for accurate and robust grape disease detection.
• To create a lightweight, mobile-deployable version for real-time, on-site agricultural monitoring.

Main Methods:

• Integration of multimodal data and parallel heterogeneous activation functions in a deep learning architecture.
• Development of a mobile-optimized model using structural pruning, quantization, and depthwise separable convolutions.
• Comparative performance analysis against established deep learning models (YOLOv3, YOLOv5, DETR, etc.).

Main Results:

• The novel model achieved 91% accuracy, 93% precision, 90% recall, and 91% mAP with 56 FPS.
• The lightweight model was successfully deployed on an iPhone 15 for real-time detection.
• Significant reduction in computational complexity and resource consumption for the mobile model.

Conclusions:

• The proposed deep learning model significantly enhances grape disease detection accuracy and robustness.
• The lightweight mobile model provides a practical solution for rapid, on-site disease identification in smart farming.
• This research contributes advanced technical solutions to smart agriculture and disease management.