From laboratory to field: cross-domain few-shot learning for crop disease identification in the field
2024

Cross-Domain Few-Shot Learning for Crop Disease Identification

Sample size: 600 publication 10 minutes Evidence: high

Author Information

Author(s): Yang Sen, Feng Quan, Zhang Jianhua, Yang Wanxia, Zhou Wenwei, Yan Wenbo

Primary Institution: Gansu Agricultural University

Hypothesis

Can cross-domain few-shot learning improve crop disease identification in field conditions?

Conclusion

The study demonstrates that cross-domain few-shot learning significantly enhances the accuracy of crop disease identification in field environments.

Supporting Evidence

  • The CDFSL-BDC model achieved an accuracy of 80.13% in the 5-shot setting.
  • The study utilized a dataset containing 9 crops and 43 disease types.
  • Cross-domain learning improved model performance significantly compared to single-domain learning.

Takeaway

This study shows how to teach computers to recognize plant diseases even when they have only seen a few examples, helping farmers identify problems faster.

Methodology

The study developed three cross-domain few-shot learning models and tested them on various datasets to evaluate their performance in identifying crop diseases.

Potential Biases

The reliance on specific datasets may introduce bias, affecting the generalization of the models to unseen domains.

Limitations

The models struggle with complex backgrounds and may misidentify diseases due to similarities with natural leaf aging or external injuries.

Digital Object Identifier (DOI)

10.3389/fpls.2024.1434222

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication