A Supervised Contrastive Learning Framework with Graph-Based Feature Extraction for Small-Sample Automatic Modulation Recognition

In this paper, we propose a supervised contrastive learning framework with graph-based feature extraction (SCL-GFE) for small-sample AMR, which includes the supervised contrastive pre-training stage and the supervised fine-tuning stage. In the first stage, the supervised contrastive loss is introduced to utilize relationships between different samples based on label information. In the second stage, the model is fine-tuned by the cross-entropy loss. Moreover, the encoder capable of extracting easily distinguishable graph-based features is simplified to prevent overfitting under the small-sample condition. Compared to cross-entropy-based methods and existing graph-based methods, experimental results justify the advantages of the proposed SCL-GFE method on recognition accuracy in the condition of a few samples with different proportions.