Multi-level Feature Representation and Multi-layered Fusion Contrast for Few-Shot Classification

Training a model that can quickly adapt to new tasks keeps a crucial challenge for few-shot learning. Approaches based on metric-learning are very popular and promising. However, the existing such approaches usually only rely on the feature information obtained in the last layer of the feature extraction backbone for similarity metric, and do not consider that the feature information obtained in multi-layered can be fully utilized. Furthermore, the targeted differentiation of features and the selection and construction of loss functions are usually ignored by these approaches, which will become important factors that limit the performance of the model. Therefore, a few-shot learning approach MFR-MFC with attention mechanism based on multi-level feature representation and multi-layered fusion contrast is proposed in this paper. First, multi-level feature representations are introduced when extracting features, feature information of multi-layers are used to perform information fusion, and then this information is utilized for subsequent metric. Then, when training the model, multi-level features are also used to introduce multi-layered fusion contrast. Additionally, an attention module is introduced in the feature extraction process to make the model obtain more discriminative information. Experiments have shown that the approach proposed in this paper has achieved excellent performance in few-shot classification and has significant advantages compared with advanced technologies.