Under Small Sample Conditions: A Communication Emitter Individual Feature Extraction Method

A focal point in the identification of individual communication emitter of the same kind lies in extracting emitter fingerprint feature vectors with robust classification capabilities to categorize individual emitter sources. Especially under small sample conditions, due to the limited labeled samples for individual entities, it's challenging to train classifiers with largescale data. This necessitates the extraction of more potent emitter source fingerprint feature vectors; otherwise, the identification accuracy would significantly decrease. Addressing this issue, we propose a feature extraction method for communication emitter under small sample conditions based on an SIB/LLE. By leveraging Locally Linear Embedding (LLE) for dimensionality reduction on the Square Integral Bispectrum (SIB), we achieve a concise feature with sufficient representational capacity, and a Support Vector Machine classifier is employed for individual identification. Experiments indicate that the proposed method achieves an identification accuracy of 76.19% while SNR is 10 and 87.08% while SNR is 20, demonstrating its superior performance in distinguishing different individuals among the same kind of communication emitter sources under small sample conditions.