TeaDiseaseNet: multi-scale self-attentive tea disease detection

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate tea disease detection is crucial for optimizing tea yield, quality, and production, while minimizing economic losses.
• Existing methods face challenges with variable disease scales and dense, obscuring disease patterns in tea images.

Purpose of the Study:

• To introduce TeaDiseaseNet, a novel deep learning method for enhanced tea disease detection.
• To address the limitations of current methods in handling complex visual characteristics of tea diseases.

Main Methods:

• Developed TeaDiseaseNet, incorporating a Convolutional Neural Network (CNN) for multi-scale feature extraction.
• Integrated a self-attention mechanism to capture global pixel dependencies and local feature interactions.
• Employed a channel attention mechanism to refine multi-scale features, reduce redundancy, and improve localization.

Main Results:

• TeaDiseaseNet demonstrated superior performance in detecting tea diseases, especially in complex backgrounds and with varying disease scales.
• Comparative experiments and ablation studies validated the effectiveness of the proposed multi-scale self-attention approach.
• The method achieved precise localization and recognition of disease information across diverse scenarios.

Conclusions:

• The proposed TeaDiseaseNet offers a significant advancement in intelligent tea disease diagnosis.
• This method has substantial potential for improving tea disease management strategies and overall production efficiency.