D. Argüeso, A. Picon, U. Irusta, A. Medela, M. G. San-Emeterio, A. Bereciartua, and A. Alvarez-Gila, “Few-Shot Learning approach for plant disease classification using images taken in the field,” Computers and Electronics in Agriculture, 2020.
Abstract:
The study used few-shot learning (FSL) algorithms with deep learning to classify plant leaf diseases using small datasets. Researchers employed the PlantVillage dataset (54,303 labeled images across 38 plant/disease types) and implemented Siamese networks with Triplet loss, fine-tuning an Inception V3 network. The approach achieved 55.5% median accuracy with only 1 image per class, 80.0% with 15 images per class, and 90.0% with 80 images per class, representing an 89.1% reduction in training data requirements with only a 4-point accuracy loss compared to full-data training. The FSL approach substantially outperformed classical fine-tuning transfer learning, demonstrating that new plant leaf and disease types can be learned with minimal labeled data.
Bibtex:
@article{argueso_few_shot_2020,
title = {Few-Shot Learning approach for plant disease classification using images taken in the field},
journal = {Computers and Electronics in Agriculture},
year = {2020},
author = {Arg{\"u}eso, D. and Picon, A. and Irusta, U. and Medela, A. and San-Emeterio, M. G. and Bereciartua, A. and {Alvarez-Gila}, Aitor},
doi = {10.1016/j.compag.2020.105542},
month = aug
}