Multi-Feature Fusion Classification Approach for Specific Communication Emitter Identification

Addressing issues such as the limited focus on features and poor interpretability in deep learning models for the individual recognition of communication emitters, this study introduces a multi-domain feature fusion model. Initially, wavelet transforms are applied to raw signals to generate two-dimensional time-frequency images. Subsequently, features are extracted from these images using neural networks to form one-dimensional image feature vectors. Additionally, multi-domain parameters of the signals are manually extracted. These two types of feature vectors are concatenated to create a combined feature set. Simulation results indicate that, on the dataset of Automatic Dependent Surveillance-Broadcast (ADS-B) emitters from multiple aircraft, selecting 10 categories of radiation sources, the proposed method improves identification accuracy by approximately 15% compared to using image features alone, achieving up to 97%. This enhancement significantly improves the accuracy of individual communication transmitter identification.