KBNet: A Language and Vision Fusion Multi-Modal Framework for Rice Disease Segmentation

Area of Science:

• Agricultural Science
• Computer Vision
• Deep Learning

Background:

• Accurate rice disease segmentation is vital for crop management.
• Existing deep learning models struggle with multi-scale and irregularly shaped lesions.

Purpose of the Study:

• To develop a novel framework for precise rice leaf disease segmentation.
• To address limitations of current methods in handling complex lesion characteristics.

Main Methods:

• Proposed KBNet, a multi-modal framework combining CNN and Transformer architectures.
• Introduced Kalman Filter Enhanced Kolmogorov-Arnold Networks (KF-KAN) for multi-scale lesion extraction.
• Developed Boundary-Constrained Physical-Information Neural Network (BC-PINN) to model irregular lesions using physical priors.

Main Results:

• KBNet demonstrated robust performance in segmenting diverse and complex rice disease patterns.
• The KF-KAN module effectively extracted and fused multi-scale lesion information.
• The BC-PINN module improved segmentation accuracy for irregular lesions and boundaries.

Conclusions:

• KBNet offers a significant advancement in rice disease segmentation technology.
• The framework provides valuable technical support for intelligent agriculture, disease identification, and management.
• KBNet shows potential for broad application in agricultural monitoring and intelligent systems.