Cross-Frequency Feature Learning for Dual-Band Radar HRRP Target Recognition

Radar ground target recognition using high-resolution range profile (HRRP) has attracted considerable attention recently. Existing methods primarily focus on single-band data, yet the HRRP of targets with similar shape structures are highly similar, making this a challenging task in complex scenarios. To address this, we use dual-band HRRPs to exploit target's scattering variations across frequencies, thereby improving recognition. In this article, a cross-frequency feature learning method is proposed to extract interfrequency discriminative features via both handcrafted and deep-learning perspectives. The handcrafted features are extracted at raw HRRP levels and their single-band feature level. The resulting heterogeneous feature sets are optimized using sparse coding and coding rate divergence maximization. For deep features, wavelet packet decomposition is integrated into a neural network to enable multiscale feature extraction. Attention mechanism is employed to characterize the wavelet domain differences between the dual-band data, with training guided by a difference search loss function. Frequency interaction further enhances the deep feature representations. Finally, deep discriminant correlation analysis is applied to jointly optimize both feature types, preserving optimal discriminative information. Extensive experiments on three datasets were conducted to verify the proposed method's effectiveness.