TY - GEN
T1 - Disentanglement Model for HRRP Target Recognition when Missing Aspects
AU - Wang, Yanhua
AU - Ma, Yunchi
AU - Zhang, Liang
AU - Wang, Junfu
AU - Zhang, Yi
AU - Lv, Hongfen
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - High resolution range profile (HRRP) is an important approach in radar automatic target recognition (RATR). Numerous deep learning methods have been proposed for HRRP recognition. However, HRRP is sensitive to aspects. When the training dataset misses aspects, the performance of the deep learning models is limited. Motivated by disentangled representation learning (DRL), this paper proposes a type-aspect disentanglement model to solve the above aspect-missing problem. The proposed method aims to obtain uncorrelated type feature and aspect feature of HRRP via DRL, and uses the type feature for recognition. Specifically, the type and aspect features are explicitly obtained via a VGG11-based network. An adversarial decorrelation loss and a reconstruction loss are applied to guide the disentanglement process. Moreover, the true type and aspect labels are employed to supervise the two features' semantic discriminability. Experimental results demonstrate the proposed method effectively improves the recognition performance in aspect-missing cases.
AB - High resolution range profile (HRRP) is an important approach in radar automatic target recognition (RATR). Numerous deep learning methods have been proposed for HRRP recognition. However, HRRP is sensitive to aspects. When the training dataset misses aspects, the performance of the deep learning models is limited. Motivated by disentangled representation learning (DRL), this paper proposes a type-aspect disentanglement model to solve the above aspect-missing problem. The proposed method aims to obtain uncorrelated type feature and aspect feature of HRRP via DRL, and uses the type feature for recognition. Specifically, the type and aspect features are explicitly obtained via a VGG11-based network. An adversarial decorrelation loss and a reconstruction loss are applied to guide the disentanglement process. Moreover, the true type and aspect labels are employed to supervise the two features' semantic discriminability. Experimental results demonstrate the proposed method effectively improves the recognition performance in aspect-missing cases.
KW - aspect-missing
KW - disentangled representation learning (DRL)
KW - High resolution range profile (HRRP)
KW - radar automatic target recognition (RATR)
UR - http://www.scopus.com/inward/record.url?scp=85178333613&partnerID=8YFLogxK
U2 - 10.1109/IGARSS52108.2023.10281604
DO - 10.1109/IGARSS52108.2023.10281604
M3 - Conference contribution
AN - SCOPUS:85178333613
T3 - International Geoscience and Remote Sensing Symposium (IGARSS)
SP - 5770
EP - 5773
BT - IGARSS 2023 - 2023 IEEE International Geoscience and Remote Sensing Symposium, Proceedings
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 2023 IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2023
Y2 - 16 July 2023 through 21 July 2023
ER -