Fusion of HRRP Time-Frequency Analysis and Multi-Scale Features for Convolutional Neural Network-Based Target Recognition

For radar target recognition in high-resolution range profiles (HRRP) under low signal-to-noise ratio (SNR) conditions, traditional methods typically involve denoising followed by recognition. However, these methods struggle with complex noise. To enhance HRRP information extraction, this paper proposes an integrated approach combining noise reduction and recognition. First, the short-time Fourier transform (STFT) is improved with a complex Gaussian window to enhance time-frequency resolution. Then, multi-scale analysis is applied by introducing scale values to better capture detailed target features. Differential operations are used to highlight scattering points and edges, improving recognition accuracy. A convolutional neural network (CNN) is employed to extract multi-level features for target recognition. Experimental results on a simulated HRRP dataset from the U.S. Air Force Research Laboratory (AFRL) demonstrate the proposed method's superior performance. It outperforms traditional methods in both accuracy and robustness, offering stronger noise resistance and better utilisation of HRRP's rich features, providing an effective solution for radar target recognition tasks.