Research on IoT Device Fingerprint Recognition in Grid Power System Based on Few-Shot Learning

IoT smart devices play a pivotal role in power grid systems, contributing to efficient monitoring and control. However, the diverse range of IoT device types and the scarcity of samples pose challenges for accurate device recognition. In this context, we propose a method tailored to small-sample scenarios, addressing the vital need for robust detection. Our approach leverages the Triplet-loss concept to construct a Triplet-CNN feature extractor, harnessing the power of CNN networks for enhanced feature extraction and achieving the construction of a privacy recognition method for IoT devices under small sample conditions. Our model demonstrates exceptional performance, achieving an impressive 84% detection accuracy with a minimal sample size of just 20, significantly surpassing traditional deep learning algorithms.