Plant pest and disease lightweight identification model by fusing tensor features and knowledge distillation

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate plant pest and disease diagnosis is crucial for crop yield and quality.
• Current methods suffer from low efficiency and accuracy due to reliance on expert experience and complex disease varieties.

Purpose of the Study:

• To develop an efficient and accurate model for plant pest and disease identification.
• To address the limitations of existing diagnostic methods in terms of speed and precision.

Main Methods:

• Proposed a Plant Pest and Disease Lightweight identification Model by fusing Tensor features and Knowledge distillation (PDLM-TK).
• Introduced Lightweight Residual Blocks based on Spatial Tensor (LRB-ST) and depth separable convolution for feature extraction and efficiency.
• Developed Branch Network Fusion with Graph Convolutional features (BNF-GC) for image segmentation and correlation feature extraction.
• Implemented a Model Training Strategy based on knowledge Distillation (MTS-KD) for balanced accuracy and efficiency.

Main Results:

• Achieved high classification accuracy (96.19%) and F1 score (94.94%) on the Plant Village dataset.
• Demonstrated superior execution efficiency compared to lightweight methods like MobileViT.
• Successfully balanced model accuracy and diagnostic efficiency.

Conclusions:

• The PDLM-TK model offers a significant advancement in automated plant pest and disease diagnosis.
• The model's efficiency and accuracy make it suitable for rapid and precise identification in agricultural settings.
• This approach provides a scalable solution for improving crop management and reducing losses.