MS-Net: a novel lightweight and precise model for plant disease identification

Area of Science:

• Computer Vision
• Agricultural Technology
• Machine Learning

Background:

• Deep learning is a primary method for plant disease identification, driven by advancements in image processing and computing power.
• Increasing neural network depth improves performance but raises computational complexity, hindering mobile deployment.
• Existing models often require extensive resources, limiting their practical application on edge devices.

Purpose of the Study:

• To develop a novel, lightweight convolutional neural network for efficient and accurate plant disease detection.
• To enhance feature representation and optimize model parameters for improved classification accuracy.
• To create a robust model suitable for deployment on mobile devices with limited computational resources.

Main Methods:

• A lightweight convolutional neural network was designed by integrating skip connections into MobileNetV3.
• An improved whale optimization algorithm was employed to optimize feature fusion weights within skip connections.
• Bias loss was utilized instead of cross-entropy loss to mitigate interference from redundant data.
• The model underwent pre-training on a plant classification dataset for enhanced performance and robustness.

Main Results:

• The proposed model achieved 99.8% accuracy on the PlantVillage dataset with significantly fewer parameters.
• It demonstrated strong performance on an apple leaf disease dataset (97.8% accuracy) with complex outdoor backgrounds.
• The model exhibited lower complexity and higher recognition accuracy compared to existing advanced plant disease diagnosis models.

Conclusions:

• The developed lightweight convolutional neural network offers a superior balance of accuracy, efficiency, and reduced complexity for plant disease detection.
• The model's design, incorporating optimized skip connections and bias loss, effectively addresses the limitations of resource-intensive deep learning approaches.
• This research provides a promising solution for deploying advanced plant disease identification systems on mobile and edge devices in real-world agricultural settings.