Improving SAR Automatic Target Recognition Models with Transfer Learning from Simulated Data

Research output: Contribution to journalJournal articleResearchpeer-review

3470 Downloads (Pure)

Abstract

Data-driven classification algorithms have proved to do well for automatic target recognition (ATR) in synthetic aperture radar (SAR) data. Collecting data sets suitable for these algorithms is a challenge in itself as it is difficult and expensive. Due to the lack of labeled data sets with real SAR images of sufficient size, simulated data play a big role in SAR ATR development, but the transferability of knowledge learned on simulated data to real data remains to be studied further. In this letter, we show the first study of Transfer Learning between a simulated data set and a set of real SAR images. The simulated data set is obtained by adding a simulated object radar reflectivity to a terrain model of individual point scatters, prior to focusing. Our results show that a Convolutional Neural Network (Convnet) pretrained on simulated data has a great advantage over a Convnet trained only on real data, especially when real data are sparse. The advantages of pretraining the models on simulated data show both in terms of faster convergence during the training phase and on the end accuracy when benchmarked on the Moving and Stationary Target Acquisition and Recognition data set. These results encourage SAR ATR development to continue the improvement of simulated data sets of greater size and complex scenarios in order to build robust algorithms for real life SAR ATR applications.
Original languageEnglish
JournalIEEE Geoscience and Remote Sensing Letters
Volume14
Issue number9
Pages (from-to)1484-8
ISSN1545-598X
DOIs
Publication statusPublished - 2017

Keywords

  • SAR ATR
  • Convolutional Neural Networks
  • Transfer Learning
  • SAR Image Simulation

Cite this

@article{2217ba7b7be1407fac61bede6e273d4e,
title = "Improving SAR Automatic Target Recognition Models with Transfer Learning from Simulated Data",
abstract = "Data-driven classification algorithms have proved to do well for automatic target recognition (ATR) in synthetic aperture radar (SAR) data. Collecting data sets suitable for these algorithms is a challenge in itself as it is difficult and expensive. Due to the lack of labeled data sets with real SAR images of sufficient size, simulated data play a big role in SAR ATR development, but the transferability of knowledge learned on simulated data to real data remains to be studied further. In this letter, we show the first study of Transfer Learning between a simulated data set and a set of real SAR images. The simulated data set is obtained by adding a simulated object radar reflectivity to a terrain model of individual point scatters, prior to focusing. Our results show that a Convolutional Neural Network (Convnet) pretrained on simulated data has a great advantage over a Convnet trained only on real data, especially when real data are sparse. The advantages of pretraining the models on simulated data show both in terms of faster convergence during the training phase and on the end accuracy when benchmarked on the Moving and Stationary Target Acquisition and Recognition data set. These results encourage SAR ATR development to continue the improvement of simulated data sets of greater size and complex scenarios in order to build robust algorithms for real life SAR ATR applications.",
keywords = "SAR ATR, Convolutional Neural Networks, Transfer Learning, SAR Image Simulation",
author = "David Malmgren-Hansen and Anders Kusk and J{\o}rgen Dall and Nielsen, {Allan Aasbjerg} and Rasmus Engholm and Henning Skriver",
year = "2017",
doi = "10.1109/LGRS.2017.2717486",
language = "English",
volume = "14",
pages = "1484--8",
journal = "I E E E Geoscience and Remote Sensing Letters",
issn = "1545-598X",
publisher = "Institute of Electrical and Electronics Engineers",
number = "9",

}

TY - JOUR

T1 - Improving SAR Automatic Target Recognition Models with Transfer Learning from Simulated Data

AU - Malmgren-Hansen, David

AU - Kusk, Anders

AU - Dall, Jørgen

AU - Nielsen, Allan Aasbjerg

AU - Engholm, Rasmus

AU - Skriver, Henning

PY - 2017

Y1 - 2017

N2 - Data-driven classification algorithms have proved to do well for automatic target recognition (ATR) in synthetic aperture radar (SAR) data. Collecting data sets suitable for these algorithms is a challenge in itself as it is difficult and expensive. Due to the lack of labeled data sets with real SAR images of sufficient size, simulated data play a big role in SAR ATR development, but the transferability of knowledge learned on simulated data to real data remains to be studied further. In this letter, we show the first study of Transfer Learning between a simulated data set and a set of real SAR images. The simulated data set is obtained by adding a simulated object radar reflectivity to a terrain model of individual point scatters, prior to focusing. Our results show that a Convolutional Neural Network (Convnet) pretrained on simulated data has a great advantage over a Convnet trained only on real data, especially when real data are sparse. The advantages of pretraining the models on simulated data show both in terms of faster convergence during the training phase and on the end accuracy when benchmarked on the Moving and Stationary Target Acquisition and Recognition data set. These results encourage SAR ATR development to continue the improvement of simulated data sets of greater size and complex scenarios in order to build robust algorithms for real life SAR ATR applications.

AB - Data-driven classification algorithms have proved to do well for automatic target recognition (ATR) in synthetic aperture radar (SAR) data. Collecting data sets suitable for these algorithms is a challenge in itself as it is difficult and expensive. Due to the lack of labeled data sets with real SAR images of sufficient size, simulated data play a big role in SAR ATR development, but the transferability of knowledge learned on simulated data to real data remains to be studied further. In this letter, we show the first study of Transfer Learning between a simulated data set and a set of real SAR images. The simulated data set is obtained by adding a simulated object radar reflectivity to a terrain model of individual point scatters, prior to focusing. Our results show that a Convolutional Neural Network (Convnet) pretrained on simulated data has a great advantage over a Convnet trained only on real data, especially when real data are sparse. The advantages of pretraining the models on simulated data show both in terms of faster convergence during the training phase and on the end accuracy when benchmarked on the Moving and Stationary Target Acquisition and Recognition data set. These results encourage SAR ATR development to continue the improvement of simulated data sets of greater size and complex scenarios in order to build robust algorithms for real life SAR ATR applications.

KW - SAR ATR

KW - Convolutional Neural Networks

KW - Transfer Learning

KW - SAR Image Simulation

U2 - 10.1109/LGRS.2017.2717486

DO - 10.1109/LGRS.2017.2717486

M3 - Journal article

VL - 14

SP - 1484

EP - 1488

JO - I E E E Geoscience and Remote Sensing Letters

JF - I E E E Geoscience and Remote Sensing Letters

SN - 1545-598X

IS - 9

ER -