Javier Areta

Synthetic aperture radar signal processing in parallel using GPGPU

In this work an efficient parallel implementation of the Chirp Scaling Algorithm for Synthetic Aperture Radar processing is presented. The architecture selected for the implementation is the general purpose graphic processing unit, as it is well suited for scientific applications and real-time implementation of algorithms. The analysis of a first implementation led to several improvements which resulted in an important speed-up. Details of the issues found are explained, and the performance improvement of their correction explicitly shown.

Weather radar data processing on graphic cards

Weather radar operation generates data at a high rate that requires prompt processing. The operations performed on data for weather product generation are repeated in each resolution cell and thus are naturally prone to parallelization. Parallel processing using graphic cards is an emerging technology that allows for implementation of high-throughput algorithms at a low cost. In this paper, the parallel implementation of the main product of a polarimetric weather radar using GPU is presented, focusing on its optimization. A speedup exceeding 20

Extended Chirp Scaling on GPGPU

\times

Polarimetric SAR Image Segmentation using CEM Algorithm

× is obtained when compared to the serial implementation. Also processing is found to be memory bound, which results in a counter-intuitive performance improvement when the number of threads per job is reduced.

Unsupervised Polarimetric SAR Image Classification Using

We propose a polarimetric SAR image classification method using the Expectation-Maximization (EM) algorithm. It is a semi-supervised algorithm with random initialization that only requires the number of clases to be identified as initial information. We apply the proposed algorithm to simulated and real Multilook Complex (MLC) polarimetric data, assuming a Gp0 mixture model. The classification performance is evaluated by means of the confusion matrix and the kappa index. Finally, we compare the results to those obtained by other authors via SEM (Stochastic EM) method using the same model and data set.

\mathcal {G}_{p}^{0}

We present in this article the implementation of Chirp Scaling Algorithm for SAR systems signal processing. This algorithm takes advantage of Chirp signals properties to process the information acquired from aerial radar systems and output high resolution images of earth ground. We propose a parallel implementation of the algorithm in GPGPU (General Purpose Graphic Processing Unit) using CUDA C. Several bottlenecks and penalizations are detected and corrected within various versions of the implementation, reducing processing time.

Mixture Model

We present the progress made in the development of an instrumentation system for fresh water bathymetry. Several design considerations are analyzed, such as transducer element characterization, SONAR equation modelling, and conceptual design of the instrumentation electronics. A Class-D, “Half-Bridge” configuration is proposed to drive the transducer used for transmission and reception. General system modules are also introduced. Future work involves the construction of a prototype based on this design to evaluate its performance.

Unsupervised classification algorithm based on EM method for polarimetric SAR images

In this work we perform Synthetic Aperture Radar (SAR) polarimetric images segmentation based on the Classification-Expectation-Maximization (CEM) method, with both supervised and unsupervised initialization. In the former case, the algorithm is randomly initialized with the number of classes as the only initial information, while in the unsupervised case initialization is based on a previous classification. Real EMISAR Single-Look-Complex (SLC) data are used, with Mixing Gaussian model. Results are compared with those obtained by Wishart unsupervised classification method, which is a well-known and widely used method for radar image classification. Finally, Davies-Bouldin index is applied for quantitative comparison between the obtained segmentations, and for studying the CEM method performance.