A method and system for classifying crop disease severity based on multiple datasets of Swin-Transformer

By improving the Swin-Transformer model and combining a deep separable residual structure and a channel attention mechanism, the problem of scarce data on small-scale crop diseases was solved, and high-accuracy classification of disease severity was achieved.

CN122223554APending Publication Date: 2026-06-16CHONGQING ACAD OF AGRI SCI
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
CHONGQING ACAD OF AGRI SCI
Filing Date
2026-04-17
Publication Date
2026-06-16

AI Technical Summary

⚠Technical Problem

Existing technologies face challenges in diagnosing diseases in small-scale specialty economic crops, including data scarcity, complexity of disease characteristics, and issues with feature scale adaptation. This makes it difficult for models to accurately classify the severity of crop diseases in real-world scenarios.

⚗Method used

An improved SwinSe classification model based on Swin-Transformer is adopted, which combines a feature extractor with a deep separable residual structure and a channel attention mechanism, and a pre-training and contrastive learning fine-tuning strategy to enhance the ability to capture local texture and edges of lesions.

🎯Benefits of technology

It significantly improved the accuracy and generalization ability of disease severity classification under conditions of scarce samples, especially achieving a high level of classification accuracy in crops such as broad beans, rice, and peppers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122223554A_ABST
    Figure CN122223554A_ABST
Patent Text Reader

Abstract

The application discloses a kind of Swin-Transformer-based multi-dataset crop disease severity classification method and system, first obtain crop disease leaf image, pre-process and data enhancement are carried out to image, and disease severity grade is marked, training set is constructed, secondly, improved network SwinSe classification model based on Swin-Transformer is constructed, then the training strategy of pre-training and contrast learning fine-tuning is used to train improved network SwinSe classification model, finally, the crop leaf image to be measured is input into the improved network SwinSe classification model trained, and the corresponding disease severity grade is output.By lightweight feature extractor, the capture ability of model to local texture and edge of disease spot is enhanced, combined with pre-training and fine-tuning strategy, the classification accuracy and generalization ability of different crop disease severity under the condition of sample scarcity are significantly improved.
Need to check novelty before this filing date? Find Prior Art