Research on Lightweight Rose Disease Detection Based on Transferable Feature Representation.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Rose leaf diseases significantly impact crop yield and quality.
• Manual disease monitoring is labor-intensive and impractical for large-scale farming.
• Deploying AI models in fields faces challenges like domain shift and limited hardware capacity.

Purpose of the Study:

• To develop an efficient and accurate method for detecting rose leaf diseases in field conditions.
• To address the performance degradation of models trained in labs when applied to real-world environments.
• To overcome the limitations of computational resources in field-deployable hardware.

Main Methods:

• Proposed a knowledge distillation approach using transferable feature representations.
• Utilized a high-capacity YOLOv12-L teacher model pre-trained on lab images.
• Trained a compact YOLOv12-N student model on field images with guidance from the teacher model.

Main Results:

• The distilled YOLOv12-N student model achieved an mAP@50 of 81.1% on the field test set.
• Demonstrated a 3.5% performance improvement over the baseline YOLOv12-N model.
• The student model maintained efficiency with only 2.56 million parameters and 6.3 GFLOPs.

Conclusions:

• Performance gains in lightweight detection primarily stem from transferring teacher model's feature representations.
• High-quality feature transfer is crucial for cross-domain agricultural vision tasks.
• The proposed method offers a generalizable and efficient solution for real-time rose leaf disease detection in precision agriculture.