With the rapid growth of Internet of Things (IoT) devices, security risks arise. Identifying device types is crucial in mitigating these risks and preventing cyber-attacks stemming from vulnerable devices. Current IoT device identification methods rely on traditional machine learning or deep learning techniques, which require extensive labeled data to create device fingerprints and require model reconstruction for each new device introduction. To overcome these limitations, we propose a few-shot learning-based approach utilizing Siamese neural networks (SNNs) for IoT device-type identification by analyzing network communications. This approach is particularly effective in scenarios with limited labeled data and constrained resources. Furthermore, it can recognize and classify previously unseen device types without necessitating model retraining. The SNN architecture is designed to learn discriminative embedding for device types by comparing device data pairs, ensuring robust performance even with limited samples. We conducted evaluations using data from real-world IoT devices. The experimental results demonstrate the proposed method’s effectiveness in accurately identifying device types with minimal labeled data, highlighting its potential to enhance network security in dynamic environments with diverse connected devices.