An HRRP ATR procedure based on deep networks is developed to replace the shallow algorithms.

Stacked Corrective Autoencoders (SCAE) is further proposed for HRRP ATR considering HRRP's characteristics.

Experiments based on measured HRRP data show the SCAE outperforms several feature extraction methods.

Some detailed experiments further validate the generalization performance of SCAE with the limit training data.