Attention Enhanced Spatial Temporal Neural Network For Hrrp Recognition

The high resolution range profile (HRRP) is an important signal for radar automatic target recognition (RATR). Recent publications have shown that exploring spatial or temporal features via neural networks is essential for this task. However, it remains a challenging problem to effectively extract and combine discriminative spatial and temporal features for HRRP recognition. In this work, we propose a novel Attention Enhanced Convolutional Gated Recurrent Unit network (AC-GRU) for HRRP recognition which improves the representation of the spatial and temporal co-occurrence in the HRRP sequences. Furthermore, an attention mechanism is employed to select key information in spatial-temporal domains. The simulation results show that the AC-GRU network can achieve better recognition rates compared with several popular classifiers under the condition of limited training data. Finally, further experiments demonstrate that our model also gets robust results under low signal-to-noise ratio.