Radar High Resolution Range Profile (HRRP), which can provide target
structure information with great potential for target recognition.
However, the structural information is not fully exploited by most
existing deep learning methods, which focus only on local or sequence
information. Furthermore, existing methods equalise target and
non-target regions in HRRP. This is not conducive to target feature
extraction. In this letter, we propose a target recognition method using
wavelet patch merging and contraction Transformer, called CT. CT can
adaptively focus on the target region and efficiently extract local and
sequence information. CT used convolution to extract local features and
contraction self-attention to extract sequential features. Wavelet patch
merging was used to avoid oversampling. Finally, the experimental
results show that the CT can effectively extract structural features in
HRRP to improve target recognition performance. It is also robust under
low signal-to-noise conditions.