Identification of Multiple Diseases in Apple Leaf Based on Optimized Lightweight Convolutional Neural Network.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate identification of apple leaf diseases is crucial for orchard management.
• Distinguishing between multiple diseases on a single leaf presents a significant challenge.
• Existing methods may struggle with complex field conditions and disease similarity.

Purpose of the Study:

• To develop an optimized RegNet model for precise identification of seven common apple leaf diseases.
• To evaluate the impact of various factors on model performance for disease classification.
• To address the challenge of disease similarity in multi-disease scenarios on apple leaves.

Main Methods:

• An optimized RegNet model was employed for apple leaf disease identification.
• Comparative analyses were performed on training methods, data expansion, optimizers, and image backgrounds.
• Offline data expansion and transfer learning with full parameter fine-tuning were utilized.

Main Results:

• Offline expansion and transfer learning enhanced model classification performance.
• Complex image backgrounds were found to significantly impact model performance.
• The optimized RegNet model achieved high testing accuracies of 93.85% and 99.23% on two datasets.

Conclusions:

• The optimized RegNet model demonstrates robust generalization for apple leaf disease identification.
• High-precision identification of multiple diseases on the same leaf is achievable under complex backgrounds.
• This approach holds significant potential for intelligent disease identification systems in apple orchards.