High-Resolution Range Profile Recognition of Midcourse Ballistic Targets Based on AMTrans

To address the problem of high-precision time-series recognition of high-resolution range profiles (HRRP) of ballistic midcourse targets, we propose a HRRP recognition method based on the multivariate time-series learning approach of Transformer encoder to further improve the recognition efficiency and accuracy. The AMTrans model is based on a multi-headed attention mechanism which can take into account multiple correlations among the input elements and adaptively adjust the attention weight in different distance units. The experimental results show that this proposed method has significant performance advantages over other methods and has better robustness.