Recognition of HRRP sequence based on TCN with attention and elastic net regularization

Traditional high-resolution range profile (HRRP) sequence recognition methods rely on artificial feature extraction and ignore the different importance of the range units within the sequence, which cannot extract deep temporal features effectively and result in poor recognition performance. This paper proposes a recognition method based on temporal convolutional network with attention and elastic net regularization (Attention-ERTCN). The proposed method extracts deep temporal information by stacking multiple TCN residual blocks. To avoid the model overfitting problem due to network deepening, we replace the weight normalization with the batch normalization in TCN to enhance the generalization ability, and introduce elastic net regularization for feature selection. Attention mechanism is introduced to adaptively assign different weights to range units, which improves the recognition ability of the target region. Average pooling is adopted to suppress the influence of local noise and preserve the global pattern of temporal features. The experimental results on MSTAR dataset show that the proposed method has better recognition performance than other temporal methods, and the rationality and effectiveness of the proposed method are verified by ablation experiments.