Improving SAR Automatic Target Recognition Models With Transfer Learning From Simulated Data

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.

[1]  Yoshua Bengio,et al.  How transferable are features in deep neural networks? , 2014, NIPS.

[2]  Qun Zhao,et al.  Support vector machines for SAR automatic target recognition , 2001 .

[3]  Tomas Pfister,et al.  Learning from Simulated and Unsupervised Images through Adversarial Training , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[4]  Christian Cochin,et al.  Classification of ships using real and simulated data in a convolutional neural network , 2016, 2016 IEEE Radar Conference (RadarConf).

[5]  Yoshua Bengio,et al.  Understanding the difficulty of training deep feedforward neural networks , 2010, AISTATS.

[6]  F. Ulaby,et al.  Handbook of radar scattering statistics for terrain , 1989 .

[7]  Jørgen Dall,et al.  Synthetic SAR Image Generation using Sensor, Terrain and Target Models , 2016 .

[8]  Michael Lee Bryant,et al.  Standard SAR ATR evaluation experiments using the MSTAR public release data set , 1998, Defense, Security, and Sensing.

[9]  Christian Cochin,et al.  Comparison of real and simulated SAR imagery of ships for use in ATR , 2010, Defense + Commercial Sensing.

[10]  Zhipeng Liu,et al.  Adaptive boosting for SAR automatic target recognition , 2007, IEEE Transactions on Aerospace and Electronic Systems.

[11]  Timo Balz,et al.  Potentials and limitations of SAR image simulators - A comparative study of three simulation approaches , 2015 .

[12]  Joseph A. O'Sullivan,et al.  SAR ATR performance using a conditionally Gaussian model , 2001 .

[13]  Lars M. H. Ulander,et al.  Synthetic-aperture radar processing using fast factorized back-projection , 2003 .

[14]  Simon Wagner.  Combination of convolutional feature extraction and support vector machines for radar ATR , 2014, 17th International Conference on Information Fusion (FUSION).

[15]  Kate Saenko,et al.  Learning Deep Object Detectors from 3D Models , 2014, 2015 IEEE International Conference on Computer Vision (ICCV).

[16]  Xian Sun,et al.  A Geometrical-Based Simulator for Target Recognition in High-Resolution SAR Images , 2012, IEEE Geoscience and Remote Sensing Letters.

[17]  David Malmgren-Hansen,et al.  Training Convolutional Neural Networks for Translational Invariance on SAR ATR , 2016 .

[18]  Raghu G. Raj,et al.  SAR Automatic Target Recognition Using Discriminative Graphical Models , 2014, IEEE Transactions on Aerospace and Electronic Systems.

[19]  David Morgan,et al.  Deep convolutional neural networks for ATR from SAR imagery , 2015, Defense + Security Symposium.

[20]  Nikola S. Subotic,et al.  Construction of hybrid templates from collected and simulated data for SAR ATR algorithms , 1998, Defense, Security, and Sensing.

[21]  Y. LeCun,et al.  Learning methods for generic object recognition with invariance to pose and lighting , 2004, Proceedings of the 2004 IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2004. CVPR 2004..