SAR COMPRESSED SENSING BASED ON GAUSSIAN PROCESS REGRESSION

The Synthetic aperture radar technology has improved in the last decades. It offers nowadays high resolution images by increasing the frequency of the modulated waves and the number of samples. This may lead to the analog to digital converter overloading, and requires high calculation time and a large storage. To overcome this problem, we propose to combine the Compressed Sensing (CS) and the Gaussian process regression (GPR) method. A part of the scene is imaged respecting the conventional acquisition, sparsified and used as a dataset for the GPR algorithm, the rest of the scene is acquired respecting the CS theory. After reconstruction, the missing pixels of the image are predicted using the model generated by the GPR algorithm. Two compression ratios and two kernel functions are evaluated. The method reconstructs an exploitable image using 40% of samples.