Loading...

Compressed Sensing in SAR

Kamjoo, Mohammad Mahdi | 2017

1456 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 50035 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Marvasti, Farokh
  7. Abstract:
  8. The remote sensing is the knowledge of gaining information about an event without having direct access to it, and synthetic aperture radars (SAR) have gained spectacular attention in this filed due to their wide applications and high efficiency. The performance of SAR, which are classified in the space-borne or space-borne radars is similar to that of pulse radars. The transmitted signals in SAR are generally chirp signals, and the received signal is two-dimensional which is scattered in two dimensions of range and azimuth called as raw data. Due to relative movement between the radar base and the target point, the distance between the radar base and the target point would not be fixed along the synthetic aperture. This results in the displacement of the data relating to the target point through the range direction. This phenomenon is called Range Cell Migration (RCM). Therefore, using the raw data as an image is not possible, and the data must be preprocessed which are called image formation algorithms. In SAR systems, the raw received signals are transmitted to the earth stations so that the processing is carried out on earth, or the entire image formation procedure is carried out inside space station, and finally, the resulted image of SAR is transmitted to earth-station. In both of mentioned scenarios of image formation, the purpose is to reduce the computational complexity and memory usage for data storage. In all of these methods, the raw data should be sampled in order to restore the image. Sampling at Nyquist rate leads to increased memory required for data process. Compressed Sensing (CS) has gained specific attention recently and is used in diverse fields. In compressed sensing, the investigated signal is sparse in a certain domain. Therefore, the signal could be reconstructed using far fewer measurements in comparison to the dimensions of the signal. In other words, the signal could be sampled at smaller sampling rates in comparison to the Nyquist rate in the non-sparse domain, and could be reconstructed in the sparse domain. In this thesis, the image restoration is carried out in the scenario where the number of target points is far fewer than the entire image samples. In other words, the restored image is sparse. Thus, instead of Nyquist sampling theorem, we can employ the compressed sensing. In order to reduce the number of measurements, we use random sampling since it has better performance in addition to maintaining simplicity in comparison to other methods. Consequently, in order to form the image using compressed sensing, we can model the problem in two formats. One method is modeling the physique of the problem as a sensing matrix, and the other is to use the inverse of traditional image processing algorithms such as Range Doppler Algorithm (RDA) and Chirp Scaling Algorithm (CSA) since these methods are linear operators and could be easily inverted. The tasks carried out in this thesis could be divided into three sections: First, traditional image formation algorithms are implemented, and the signals are reconstructed using these algorithms. Second: using the physique of problem, and invertion of image reconstruction process as the sensing matrix and restoration using sparse methods. As the simulations revealed, the sensing matrix of the problem physique has better performance, and the sampling rate could be reduced further in these methods but it is so time-consuming than other method, because in that model we can use Fast Fourier Transform (FFT) and some matrix multiplication instead of restore the matrix. Third: A new reconstruction method based on ML, MLBM estimation, which shows improved performance in comparison to other methods taking into account the complexities this method adds to the problem
  9. Keywords:
  10. Random Sampling ; Range Cell Migration Correction (RCMC) ; Chirp Signals ; Synthetic Aperture Radar (SAR) ; Compressive Sensing ; Remote Sensing ; Range Cell Migration Correction (RCMC)

 Digital Object List

 Bookmark

...see more