ViTKAB: an efficient deep learning network for cotton leaf disease identification.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Cotton is a crucial global crop, but its yield and quality are significantly impacted by leaf diseases.
• Common cotton leaf diseases include brown spot, verticillium wilt, wheel spot, and fusarium wilt, posing economic threats.

Purpose of the Study:

• To develop an advanced AI model for accurate and efficient cotton leaf disease recognition.
• To improve the robustness and inference speed of disease detection systems for agricultural applications.

Main Methods:

• Proposed ViTKAB, a novel cotton disease recognition model utilizing an enhanced Vision Transformer.
• Integrated a Kolmogorov-Arnold network and a BiFormer module into the Vision Transformer architecture.
• Optimized the model for nonlinear feature representation and sparse dynamic attention to enhance accuracy and speed.

Main Results:

• ViTKAB achieved an average recognition accuracy of 98.05% across four common cotton leaf diseases.
• The model demonstrated superior performance compared to established models like CoAtNet-7, CLIP, and PaLI.
• The enhanced Vision Transformer architecture improved inference speed and captured complex disease characteristics effectively.

Conclusions:

• The ViTKAB model offers significant advancements in intelligent crop disease detection.
• The method shows strong potential for practical deployment on edge devices in agricultural settings.
• This research provides valuable insights for developing next-generation AI-powered agricultural monitoring systems.