Rice Disease Detection: TLI-YOLO Innovative Approach for Enhanced Detection and Mobile Compatibility

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Rice disease detection is crucial for global food security and sustainable agriculture.
• Existing techniques suffer from low efficiency, poor accuracy, and high data requirements, limiting mobile deployment.
• There is a need for accurate, resource-efficient rice disease identification models suitable for mobile devices.

Purpose of the Study:

• To develop a highly accurate and resource-efficient rice disease detection model for mobile deployment.
• To address the limitations of existing rice disease identification technologies.

Main Methods:

• Proposed the Transfer Layer iRMB-YOLOv8 (TLI-YOLO) model, modifying YOLOv8 with transfer learning.
• Integrated a novel small object detection layer and the iRMB attention mechanism (Inverted Residual Blocks and Transformers) with deep separable convolution.
• Utilized the WIoUv3 loss function with a dynamic non-monotonic aggregation mechanism for improved bounding box evaluation.

Main Results:

• The TLI-YOLO model achieved 93.1% precision, 88% recall, 95% mAP, and 90.48% F1 score with 12.60 GFLOPS.
• Demonstrated significant improvements over YOLOv8n: +7.8% precision, +7.2% recall, +7.6% mAP@.5.
• Achieved real-time detection (30 FPS) on an Android device, meeting on-site diagnosis needs.

Conclusions:

• The TLI-YOLO model offers a robust and efficient solution for rice disease detection.
• Its mobile compatibility and improved performance support practical on-site agricultural monitoring.
• This technology provides vital support for enhancing rice production and sustainable agricultural practices.