This survey comprehensively evaluates the underlying mechanisms, inherent strengths, and specific weaknesses of each FSIC methods into three categories: Metric Learning, Optimization/Meta-Learning, and Transfer Learning with Large Model Fine-Tuning.
Abstract
While deep learning has achieved remarkable success in image recognition, it relies heavily on massive amounts of labeled data. In real-world scenarios like medical imaging, collecting such large datasets is often expensive or impossible. To overcome this critical bottleneck, Few-Shot Image Classification (FSIC) has emerged as a vital domain, enabling models to learn new categories from extremely limited samples. Based on a comprehensive review of recent literature, this paper systematically divides mainstream FSIC methods into three categories: Metric Learning, Optimization/Meta-Learning, and Transfer Learning with Large Model Fine-Tuning. This survey comprehensively evaluates the underlying mechanisms, inherent strengths, and specific weaknesses of each category. Additionally, the paper provides a clear performance comparison to illustrate the evolutionary impact of these different approaches. Finally, the paper deeply discusses remaining technical challenges, specifically detailing issues like cross-domain generalization and computational dependence, and outlines promising future research directions, such as multimodal knowledge fusion, to further advance the field.
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...
Huie Zhang, Mary Jane C. Samontet· International journal of com...· 0 citations
Few-Shot Image Classification (FSL) aims to identify and classify new image categories using only a very small number of labeled samples. Addressing the shortcomings of existing meta-learning and metric learning methods, such as insufficient generalization of feature representations with very few samples, overfitting,...
Xua Li, Ru-Jun Wang· International Conference on...· 0 citations
An Auxiliary Supervision-guided Edge Prototype Purification Network (EPPNet) is proposed, which introduces a novel Prototype Purification Module (PPM) and selectively integrates high-confidence edge prototypes via a learnable adaptive threshold, aiming to substantially alleviate boundary feature contamination.
Wen-Jie Meng, Kai Liu, Minghui Wang· IEEE Transactions on Medical...· 0 citations
A lightweight hierarchical representation learning framework is proposed with three key innovations that outperforms state-of-the-art methods across different settings and achieves a classification accuracy of 59.05% on EuroSAT 1-shot and 34.78% on ISIC 1-shot.
Lei-Lei Ding, Shu-Liang Zhao· International Journal of Mac...· 0 citations
A comprehensive review of recent advances in deep learning for few-shot HSI classification, systematically structured around core architectures, learning paradigms, and rigorous benchmarking to eliminate pervasive cross-study evaluation biases.
Qi-Yuan Wang, Shuo Zhang, Fei-Fei Yang et al.· IEEE Journal of Selected Top...· 1 citation
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.