基于时频增强的小样本跳频信号调制识别OA

To address the performance degradation of frequency-hopping(FH)signal modulation recognition under small-sample conditions,this paper proposes a time-frequency transformation-based data augmentation method for FH signal modulation recognition.Focusing on single-hop time-frequency grayscale images of FH signals,the proposed method employs three categories of time-frequency augmentation strategies,namely discrete wavelet transform(DWT),short-time Fourier transform(STFT),and low-energy region perturbation,to decompose,perturb,and reconstruct the detailed components of time-frequency images.In this manner,structurally diverse augmented samples are generated,ena-bling effective expansion of the training dataset.Specifically,the DWT-and STFT-based augmentation methods perturb and reconstruct the time-frequency image details through all-zero replacement,random-zero replacement,and random-noise replacement strategies,thereby enhan-cing sample diversity.Simulation results demonstrate that,under small-sample conditions,the proposed time-frequency augmentation method can significantly improve the modulation recognition performance of FH signals,achieving an overall accuracy improvement of more than 13％compared with using only the original samples,and exhibiting superior effectiveness relative to conventional data augmentation methods.