Remote Sensed Image Analysis
And Automatic Target Recognition

SARATR


ABSTRACT

Remote sensed imaging, such as synthetic aperture radar (SAR), infra-red (IR), hyperspectral etc. has seen vast improvements in the last decade due to advances in sensors and capture technology. Likewise, sophisticated image analysis and classification techniques may now be used to mine objects of interests in these images. Among the most promising methods for object, e.g. target, detection in remote sensed imagery, sparse representation based methods have shown an impressive resiliency to noise, blur, and occlusion. However, determining the best features and lack of training data continue to be outstanding open challenges in ATR. Our research combines sparsity based methods with dictionary learning and graphical priors which are shown to be very suitable for recognition tasks in presence of noise and to learn meaningful features from raw data. We advocate using class-specific priors or constraints to capture structure on sparse coefficients that helps explicitly distinguish between target classes while significantly lessening the training burden. Successful results have been shown for benchmark SAR, hyperspectral and mid-wave infra-red (MWIR) image databases.

Database

MSTAR benchmark - Example images of MSTAR database.

Sparse Representation-based Classification

Sparse representation-based classification using spike-and-slab priors. Shown in the figure are two classes of SAR images.

Classification error vs. training sample size. Proposed IGT method is compared with EMACH filter [1], SVM [2], conditionally Gaussian model [3], and AdaBoost method [4]. (Left) EOC-1. (Right) EOC-2.

Related Publications

  1. Vu H, Tiep, Nguyen L H, and Vishal Monga. "Classifying Multichannel UWB SAR Imagery via Tensor Sparsity Learning Techniques." IEEE Transactions on Aerospace and Electronic Systems 55.4 (2019): 1712-1724.

  2. Srinivas, Umamahesh, Vishal Monga, and Raghu G. Raj. "SAR automatic target recognition using discriminative graphical models." IEEE transactions on aerospace and electronic systems 50.1 (2014): 591-606.

  3. Srinivas, Umamahesh, et al. "Exploiting sparsity in hyperspectral image classification via graphical models." IEEE Geoscience and Remote Sensing Letters 10.3 (2013): 505-509.

  4. Srinivas, Umamahesh, et al. "Discriminative graphical models for sparsity-based hyperspectral target detection." Geoscience and Remote Sensing Symposium (IGARSS), 2012 IEEE International. IEEE, 2012.

  5. Srinivas, Umamahesh, et al. "Structured sparse priors for image classification." IEEE Transactions on Image Processing 24.6 (2015): 1763-1776.

Selected References

  1. Singh, R. and Kumar, B. V. Performance of the extended maximum average correlation height (EMACH) filter and the polynomial distance classifier correlation filter (PDCCF) for multi-class SAR detection and classification. Proceedings of SPIE, vol. 4727, Algorithms for Synthetic Aperture Radar Imagery IX, 2002, pp. 265–276

  2. Zhao, Q. and Principe, J. Support vector machines for SAR automatic target recognition. IEEE Transactions on Aerospace and Electronic Systems, 37, 2 (Apr. 2001), 643–654.

  3. O’Sullivan, J. A. et al. SAR ATR performance using a conditionally Gaussian model. IEEE Transactions on Aerospace and Electronic Systems, 37, 1 (2001), 91–108.

  4. Sun, Y. et al. Adaptive boosting for SAR automatic target recognition. IEEE Transactions on Aerospace and Electronic Systems, 43, 1 (Jan. 2007), 112–125.

Email
ipal.psu@gmail.com

Address
104 Electrical Engineering East,
University Park, PA 16802, USA

Lab Phone:
814-863-7810
814-867-4564