Enhanced few-shot learning for plant leaf diseases recognition

Abstract

With the breakthrough progress of deep learning technology in multiple fields, its application in specialized areas such as plant leaf disease recognition is constrained by the cost of data annotation and the lack of sample diversity. This study proposes an enhanced few-shot learning method that integrates self-supervised learning and semi-supervised learning to improve the model's generalization ability in plant leaf disease recognition tasks. Through self-supervised pre-training and semi-supervised fine-tuning, the model can effectively utilize limited annotated data and expand the training set by generating high-quality pseudo-labels. Experimental results show that this method significantly improves the model's recognition performance on unseen categories. Future research will explore more self-supervised tasks and complex pseudo-label generation algorithms to further enhance the model's accuracy and robustness, promoting the application of few-shot learning technology in the field of agriculture.

Code and data availability

The paper uses the Plant Village dataset, but it is cited prior work ([4]) with no authors' public URL, deposit, or availability language for any paper-specific data, code, models, or supplements. No qualifying assets are present in the supplied blocks.

No evidence-backed public reproduction asset is currently recorded.