Corn Disease Detection Using Few-Shot Learning Prototypical Network

The timely and accurate detection of plant diseases is crucial for minimizing crop losses and improving food security. Despite the ample adoption of deep learning in the plant disease-detection field, this approach relies heavily on a large number of datasets for successful training and classification. Recently, these challenges have been addressed by few-shot learning. In this paper, a prototypical network based on a metric learning model for the detection of corn diseases is proposed. The model uses ResNet-50 as the backbone network. It supports optimization in terms of feature extraction and classification. Thus, even if the model is trained with fewer examples, it provides good accuracy in agriculture. In the proposed work, under Four-way Five-shot and Four-way Three-shot approaches, the model exhibits notable accuracy at 97.30% and 93%, respectively.