High-Precision Tomato Disease Detection Using NanoSegmenter Based on Transformer and Lightweighting.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• The increasing application of AI and deep learning in agriculture highlights the need for advanced plant disease detection methods.
• Tomato diseases pose significant threats to agricultural economic benefits and food safety, necessitating high-precision detection solutions.

Purpose of the Study:

• To develop a high-precision detection model for tomato diseases.
• To optimize the model for performance and computational efficiency in agricultural settings.

Main Methods:

• Construction of a dedicated tomato disease image dataset.
• Proposal of a NanoSegmenter model utilizing Transformer architecture.
• Integration of lightweight technologies: inverted bottleneck, quantization, and sparse attention.

Main Results:

• The NanoSegmenter model achieved high performance metrics: 0.98 precision, 0.97 recall, and 0.95 mIoU.
• The model demonstrated significant computational efficiency with an inference speed of 37 FPS.
• Experimental validation confirmed the model's effectiveness in tomato disease detection.

Conclusions:

• The study presents an effective AI-driven solution for high-precision tomato disease detection.
• The optimized NanoSegmenter model offers valuable insights and a reference for future agricultural AI research.