A Classification Recognition Method of UAVs and Birds Based on a Fusion Measure with the VMD-Driven Model and Time–Frequency Spectrum Segmentation

To improve the classification and recognition ability of radar detection unmanned aerial vehicles (UAV) under low signal-to-noise ratio conditions, this paper propose a two-domain feature extraction method that combines variational mode decomposition (VMD)-based modal features with short-time Fourier transform (STFT)-based subband statistical features. The modal energy ratio and subband mean are selected as representative features from the two domains and are normalized and fused to construct the final feature vector. A weighted Gaussian Naive Bayes classifier is then proposed for UAV–bird discrimination. The feature weights are estimated from the training data, and the final classification decision is obtained using the maximum a posteriori criterion. Experimental results under different low-SNR conditions show that the proposed method maintains competitive classification performance and retains useful discriminative information under strong noise interference. The results indicate that the combination of VMD modal features and STFT subband statistical features provides an effective representation for UAV–bird classification under low-SNR conditions.