Few-shot image classification algorithm based on deep learning and feature fusion
Few-shot image classification remains difficult because a model must identify novel classes from only one or a few labeled examples while preserving discriminative local information. Metric-learning methods based on Earth Mover’s Distance (EMD) improve local correspondence by representing an image as a set of regional...