Automatic target recognition (ATR) from radar images remains a key challenge as imaging radar systems become more sophisticated and the electromagnetic spectrum becomes more congested. In particular, interrupted dwells on the target can cause gaps in the azimuth frequency domain and notching to avoid radio frequency interference can cause gaps in the range frequency domain. Similar problems can arise in the use of multistatic synthetic aperture radar where coverage of K-space may not be complete but is more likely to consist of a number of noncontiguous patches. One approach to dealing with such fragmented K-space is to use Compressive Sensing (CS) reconstruction techniques. This paper will assess the utility of CS reconstructions and make comparisons with the naïve Fourier reconstruction. This assessment will be made in terms of the robustness of classification performance obtained using both convolutional neural networks (CNNs) and feature-based approaches.
[1]
Michael P. Friedlander,et al.
Probing the Pareto Frontier for Basis Pursuit Solutions
,
2008,
SIAM J. Sci. Comput..
[2]
Heekuck Oh,et al.
Neural Networks for Pattern Recognition
,
1993,
Adv. Comput..
[3]
Timothy D. Ross,et al.
Adaptive SAR ATR problem set (AdaptSAPS)
,
2004,
SPIE Defense + Commercial Sensing.
[4]
Rasmus Berg Palm,et al.
Prediction as a candidate for learning deep hierarchical models of data
,
2012
.
[5]
Lawrence D. Jackel,et al.
Backpropagation Applied to Handwritten Zip Code Recognition
,
1989,
Neural Computation.