Cross-Domain Few-Shot Learning for Crop Disease Identification
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)
Want to read the original?
Access the complete publication on the publisher's website