Deep Discriminative Dictionary Learning for HRRP-based Ground Target Recognition

Traditional dictionary learning algorithms always focus on minimizing reconstruction error and learning "bias" and "features" by matrix factorization, while the emphasis of the millimeter-wave HRRP recognition is to reconstruct the discriminative representations. A discriminative dictionary learning algorithm is proposed to enhance the effectiveness of feature extraction, which involves the penalty term of the interclass reconstruction cost. Following the great success of deep learning, a deep dictionary learning model is further designed to capture the internal structural relationship of the HRRP samples. Experimental results show that the deep reconstructive dictionary learning model could achieve robust and superior classification accuracies while others do not yield at-par performance.