TFDNET: Time-Frequency Directional Enhancement Network for UAV and Bird Micro-Doppler Recognition Under Limited Sensing Symbols

Abstract

Reliable recognition between unmanned aerial vehicles (UAVs) and birds is essential for low-altitude surveillance in integrated sensing and communication (ISAC) systems. In practical orthogonal frequency division multiplexing (OFDM)-ISAC systems, frame configurations limit sensing symbols, which degrades micro-Doppler spectrograms and complicates UAV and bird recognition. To address this problem, this paper proposes a recognition method based on a Time-Frequency Directional Enhancement Network (TFDNet) for UAV and bird micro-Doppler recognition under limited sensing symbols. Initially, continuous echo scattering models are established for UAVs and birds, and short-time Fourier transform (STFT) is then adopted to generate micro-Doppler spectrograms. Based on a convolutional backbone, TFDNet integrates a Directional Convolution Block (DCB) and a Time-Frequency Profile Module (TFPM) to extract more discriminative time-frequency features for classification. Evaluation on a four-class low-altitude target recognition task under limited sensing symbols shows that TFDNet outperforms the baseline and representative comparison models.