Estudio e implementación en unidades de procesamiento gráficos GPU S de un formador de imagen SAR usando técnica de backprojection / Study and implementation on graphics processing units GPUs for SAR image formation using the backprojection algorithm

Guevara Díaz, Sergio A. (2020) Estudio e implementación en unidades de procesamiento gráficos GPU S de un formador de imagen SAR usando técnica de backprojection / Study and implementation on graphics processing units GPUs for SAR image formation using the backprojection algorithm. Proyecto Integrador Ingeniería en Telecomunicaciones, Universidad Nacional de Cuyo, Instituto Balseiro.

[img]
Vista previa
PDF (Tesis)
Español
20Mb

Resumen en español

El algoritmo Backprojection en el domino temporal fue presentado como una simple pero efectiva técnica para enfocar imágenes SAR sin hacer ninguna suposición, siendo capaz de procesar datos de cualquier modo de adquisición SAR y realizar compensación de movimiento de forma natural en el mismo algoritmo. Su principal desventaja es su carga computacional de O(N"3), convirtiéndolo en una opción lenta y no conveniente para implementar en procesamiento serial. Sin embargo, dado el constante avance tecnológico de las unidades de procesamiento gráfico de propósito general (GPGPU) que permite programación en paralelo, ahora es posible considerar la inclusión del Backprojection en el conjunto típico de algoritmos de formación de imágenes SAR. Este proyecto se enfoca en entender los fundamentos de las adquisiciones SAR y en la implementación del algoritmo Backprojection en una GPU NVIDIA T4 para procesamiento en tierra y en tiempo real. En este documento se presenta de forma gráfica e intuitiva las bases teóricas del Backprojection. Dos métodos de interpolación en rango fueron propuestos, el primero es un sobremuestreo de los ecos recibidos y el segundo es un interpolador Sinc. Con ambos métodos se logro reducir los lóbulos secundarios de la respuesta en acimut. Se encontró que si la cantidad de memoria de la GPU no es un factor limitante entonces efectuar un sobremuestreo es mas eficiente, de no ser así, el interpolador Sinc logra su cometido a cambio de un tiempo de procesamiento superior. Se aplico una ventana Kaiser de parámetro β = 2,5 en la compresión en rango, logrando reducir los lóbulos secundarios de la respuesta desde una PSLR de -13 dB hasta -20 dB. Se integro una rutina de compensación de movimiento en el mismo Backprojection y exitosamente se lograron corregir errores de enfoque debido a trayectorias no ideales de la plataforma SAR. Se implemento con la librera OpenCV un post procesamiento de imágenes dedicado a corregir deformaciones geométricas debido a adquisiciones con squint y a espaciamientos en rangos no constantes, además de esto se aplico una ecualización de histograma adaptativo de contraste limitado y un ltro de mediana para reducir speckle noise del producto nal. Finalmente, se presenta la implementación de un procesador en tiempo real Backprojection. Para lograr la velocidad de procesamiento requerida, la versión básica del algoritmo sin la interpolación ni compensation de movimiento fue ejecutada. Para probar esta rutina se realizo una simulación y se obtuvieron resultados satisfactorios, sin embargo aun se necesita una interfaz real con instrumentos SAR. El algoritmo fue escrito en Python y CUDA/C usando la librera CuPy, Para probar el algoritmo se simularon bancos puntuales con el simulador de INVAP, SIDRA y también se usaron datos reales del proyecto SARAT proporcionados por CONAE.

Resumen en inglés

Time domain Backprojection was presented as a basic yet effective algorithm to focus SAR images, being able to process data from any type of SAR imaging mode and perform motion compensation naturally in the same algorithm. The downside of this technique is it's O(N"3) computational complexity, making it a slow non convenient option to perform in serial processing. However, due to the constantly increasing technology of General-Purpose Graphics Processing Units (GPGPU), which allows parallel programming, it is now possible to consider the inclusion of the Backprojection in the every day SAR image formation algorithms. This project focusses on understanding the basics of SAR data acquisitions and in the implementation of the Backprojection algorithm on a GPU NVIDIA T4 for offline and real time data processing. In this document, an intuitive and graphical interpretation of the Backprojection's theory is presented. Two methods of signal interpolation in range were proposed, the first one is the digital upsampling of the echoed signals and the second one is a Sinc interpolator. Both methods achieved a reduction in the secondary lobes of the acimut response. It was found that if GPU memory storage is not a limiting factor then performing an upsampling is more ecient, if thats not the case, then the Sinc interpolation can do the job at the expense of an increased processing time. The kaiser window with parameter β = 2,5 was applied on the range compression, suppressing secondary lobes from the range response from a PSLR of -13 dB to -20 dB. A motion compensation routine integrated on the same Backprojection algorithm is presented, successfully correcting focusing errors given a non ideal motion of the SAR platform. An image post processing processing dedicated to correct geometrical deformations such as squinted geometries and non constant range steps was implemented using the OpenCV library, as well as a Contrast Limited Adaptive Histogram Equalization and a median filter to reduce the speckle noise on the nal product. Finally, an implementation of a real time Backprojection processing was presented. In order to achieve the processing speed needed, the basic version of the Backprojection algorithm without interpolation nor motion compensation was executed. To test this routine a simulation was carried over with successful results, nevertheless a real interface with external SAR instruments is still required. The algorithm is written on Python and CUDA/C using the CuPy library. To test the algorithm, simulations of point targets were done with the INVAP's SIDRA simulator and also, real data from the SARAT project provided from CONAE was used. The real data was taken in between Cordoba and Falda del Carmen, Argentina.

Tipo de objeto:Tesis (Proyecto Integrador Ingeniería en Telecomunicaciones)
Palabras Clave:Radar; [Backprojection algorithm; Algoritmo backprojection; SAR imaging; Imágenes SAR]
Referencias:[1] Sanders, J., Kandrot, E. CUDA by example: an introduction to general-purpose GPU programming. Addison-Wesley Professional, 2010. ix, ix, 6, 7, 8, 9 [2] Documentation, C. T. v7. 5, nvidia, mar 2016. ix, 8, 9, 10 [3] Richards, M. A., Scheer, J. A., Holm, W. A., Beckley, B., Mark, P., Richards, A., et al. Principles of modern radar volume I-basic principles. Citeseer, 2010. 1, 71 [4] Ozdemir, C. Inverse synthetic aperture radar imaging with MATLAB algorithms, tomo 210. John Wiley & Sons, 2012. 2, 12, 17, 28 [5] Duersch, M. I. Backprojection for synthetic aperture radar, 2013. 3 [6] Malaya, N., Che, S., Greathouse, J. L., Van Oostrum, R., Schulte, M. J. Accelerating matrix processing with gpus. En: 2017 IEEE 24th Symposium on Computer Arithmetic (ARITH), pags. 139{141. IEEE, 2017. 6 [7] Taher, M. Accelerating scientic applications using gpu's. En: 2009 4th International Design and Test Workshop (IDT), pags. 1{6. IEEE, 2009. 6 [8] Steinkraus, D., Buck, I., Simard, P. Using gpus for machine learning algorithms. En: Eighth International Conference on Document Analysis and Recognition (ICDAR'05), pags. 1115{1120. IEEE, 2005. 6 [9] Farber, R. CUDA application design and development. Elsevier, 2011. 6 [10] Soyata, T. GPU parallel program development using CUDA. CRC Press, 2018. 38 [11] Storti, D., Yurtoglu, M. CUDA for engineers: an introduction to high-performance parallel computing. Addison-Wesley Professional, 2015. [12] Cheng, J., Grossman, M., McKercher, T. Professional CUDA c programming. John Wiley & Sons, 2014. 6 [13] Harris, M. An easy introduction to cuda c and c++, 2012. 6 [14] NVIDIA. NVIDIA T4 TENSOR CORE GPU, 2019. URL https: //www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/tesla-t4/ t4-tensor-core-datasheet-951643.pdf. 9, 39 [15] inc, P. N., inc, P. I. Cupy documentation, release 8.0.0b3. 10 [16] Cumming, I. G., Wong, F. H.-c. Digital processing of synthetic aperture radar data: algorithms and implementation. Artech house Norwood, MA, 2005. 17, 19, 20, 26, 28, 29, 30, 31 [17] Yu, T., Li, W. A high squint airborne sar imaging algorithm with low prf. En: 2015 IEEE China Summit and International Conference on Signal and Information Processing (ChinaSIP), pags. 1076{1080. IEEE, 2015. 21 [18] Zhang, L., Sheng, J., Xing, M., Qiao, Z., Xiong, T., Bao, Z. Wavenumber-domain autofocusing for highly squinted uav sar imagery. IEEE Sensors Journal, 12 (5), 1574{1588, 2011. 21 [19] Wang, S., Wang, S., Liu, W. High-resolution squinted sar imaging algorithm applied in the near-eld environment. Complexity, 2020, 2020. 21 [20] Davidson, G. W., Cumming, I. Signal properties of spaceborne squint-mode sar. IEEE Transactions on Geoscience and Remote Sensing, 35 (3), 611{617, 1997. 23 [21] Gu, F.-f., Zhang, Q., Yan, J.-b., Jiang, H., Li, Y. A novel imaging method for highly squinted sar based on compressed sensing. En: 2015 IEEE 5th Asia-Pacic Conference on Synthetic Aperture Radar (APSAR), pags. 250{253. IEEE, 2015. 23 [22] Sommer, A., Ostermann, J. Explicit motion compensation for backprojection in spotlight sar. En: 2016 17th International Radar Symposium (IRS), pags. 1{4. IEEE, 2016. 23 [23] Levanon, N., Mozeson, E. Radar signals. John Wiley & Sons, 2004. 25 [24] Thomas, G., Flores, B. C., Sok-Son, J. Sar sidelobe apodization using the kaiser window. En: Proceedings 2000 International Conference on Image Processing (Cat. No. 00CH37101), tomo 1, pags. 709{712. IEEE, 2000. 28 [25] Cumming, I., Neo, Y., Wong, F. Interpretations of the omega-k algorithm and comparisons with other algorithms. En: IGARSS 2003. 2003 IEEE International Geoscience and Remote Sensing Symposium. Proceedings (IEEE Cat. No. 03CH37477), tomo 3, pags. 1455{1458. IEEE, 2003. 30 [26] Sommer, A., Nguyen, M. P., Ostermann, J. Comparison of omega-k and backprojection regarding spatial resolution for squinted spotlight sar with motion errors. En: 2015 IEEE 5th Asia-Pacic Conference on Synthetic Aperture Radar (APSAR), pags. 143{147. IEEE, 2015. 31 [27] Melvin, W. L., Scheer, J. A. Principles of Modern Radar Vol. II: Advanced Techniques, 2013 by SciTech Publishing. 35, 71 [28] Vlothuizen, W. J., Ditzel, M. Real-time brute force sar processing. En: 2009 IEEE Radar Conference, pags. 1{4. IEEE, 2009. 36 [29] Moon, K., Long, D. G. A new factorized backprojection algorithm for stripmap synthetic aperture radar, 2013. 36 [30] Hartley, T. D., Fasih, A. R., Berdanier, C. A., Ozguner, F., Catalyurek, U. V. Investigating the use of gpu-accelerated nodes for sar image formation. En: 2009 IEEE International Conference on Cluster Computing and Workshops, pags. 1{8. IEEE, 2009. 36 [31] Yaroslavsky, L. Fast discrete sinc-interpolation: a gold standard for image resampling. Advances in Signal Transforms: Theory and Applications, 7, 337, 2007. 43, 46 [32] Haykin, S., Van Veen, B. Signals and systems. John Wiley & Sons, 2007. 46 [33] Alexander Mordvintsev, A. K. Opencv - python tutorials documentation release 1. 54, 56 [34] Schafer, R. W. What is a savitzky-golay lter?[lecture notes]. IEEE Signal processing magazine, 28 (4), 111{117, 2011. 61 [35] Gallagher, N. B. Savitzky-golay smoothing and dierentiation lter. 61 [36] Wahl, D. E., Eichel, P., Ghiglia, D., Jakowatz, C. Phase gradient autofocusa robust tool for high resolution sar phase correction. IEEE Transactions on Aerospace and electronic systems, 30 (3), 827{835, 1994. 65 [37] Samczynski, P., Kulpa, K. S. Coherent mapdrift technique. IEEE Transactions on Geoscience and Remote Sensing, 48 (3), 1505{1517, 2009. 65 [38] Lim, T. S., Koo, V. C., Ewe, H.-T., Chuah, H.-T. A sar autofocus algorithm based on particle swarm optimization. Progress In Electromagnetics Research, 1, 159{176, 2008. 65 [39] Stringham, C., Long, D. G. Gpu processing for uas-based lfm-cw stripmap sar. Photogrammetric Engineering & Remote Sensing, 80 (12), 1107{1115, 2014. 68 [40] Guevara, S. Backprojection online processing by blocks - w552t21808753, 2020. URL https://youtu.be/O89nvxP1vBs. 69 [41] Guevara, S. Backprojection online processing by blocks - w552t14758669, 2020. URL https://youtu.be/MOeZMZFZSP8. 69 [42] Soumekh, M. Synthetic aperture radar signal processing, tomo 7. New York: Wiley, 1999. 71 [43] Zhang, H., Tang, J., Wang, R., Deng, Y., Wang, W., Li, N. An accelerated backprojection algorithm for monostatic and bistatic sar processing. Remote Sensing, 10 (1), 140, 2018. 76 [44] Rasmussen, A. C. Implementation and performance of factorized backprojection on low-cost commercial-o-the-shelf hardware, 2016. [45] Ulander, L. M., Hellsten, H., Stenstrom, G. Synthetic-aperture radar processing using fast factorized back-projection. IEEE Transactions on Aerospace and electronic systems, 39 (3), 760{776, 2003. [46] Moon, K., Long, D. G. A new factorized backprojection algorithm for stripmap synthetic aperture radar, 2013. [47] Ran, L., Liu, Z., Xie, R., Zhang, L. Focusing high-squint synthetic aperture radar data based on factorized back-projection and precise spectrum fusion. Remote Sensing, 11 (24), 2885, 2019. [48] Rogan, A., Carande, R. Improving the fast back projection algorithm through massive parallelizations. En: Radar Sensor Technology XIV, tomo 7669, pag. 76690I. International Society for Optics and Photonics, 2010. 76 [49] Nguyen, L., Ressler, M., Wong, D., Soumekh, M. Enhancement of backprojection sar imagery using digital spotlighting preprocessing. En: Proceedings of the 2004 IEEE Radar Conference (IEEE Cat. No. 04CH37509), pags. 53{58. IEEE, 2004. 76 [50] Desai, M. D., Jenkins, W. K. Convolution backprojection image reconstruction for spotlight mode synthetic aperture radar. IEEE Transactions on Image Processing, 1 (4), 505{517, 1992. [51] Wahl, D. E., Yocky, D. A., Jakowatz Jr, C. V. An implementation of a fast backprojection image formation algorithm for spotlight-mode sar. En: Algorithms for Synthetic Aperture Radar Imagery XV, tomo 6970, pag. 69700H. International Society for Optics and Photonics, 2008. [52] Shen, S., Nie, X., Zhang, X. Research on synthetic aperture radar processing for the spaceborne sliding spotlight mode. Sensors, 18 (2), 455, 2018. 76 [53] Capozzoli, A., Curcio, C., Liseno, A. Fast gpu-based interpolation for sar backprojection. Progress In Electromagnetics Research, 133, 259{283, 2013. 76 [54] Ash, J. N. An autofocus method for backprojection imagery in synthetic aperture radar. IEEE Geoscience and Remote Sensing Letters, 9 (1), 104{108, 2011. 76 [55] Guo, Z., Zhang, H., Ye, S. Simplied and approximation autofocus back-projection algorithm for sar. The Journal of Engineering, 2019 (20), 6408{6412, 2019. [56] Duersch, M. I., Long, D. G. Backprojection autofocus for synthetic aperture radar, 2013. [57] Saeedi, J., Faez, K. A back-projection autofocus algorithm based on ight trajectory optimization for synthetic aperture radar imaging. Multidimensional Systems and Signal Processing, 27 (2), 411{431, 2016. 76
Materias:Ingeniería en telecomunicaciones > Procesamiento digital de señales
Divisiones:Gcia. de área de Investigación y aplicaciones no nucleares > Laboratorio de investigación aplicada en Telecomunicaciones
Código ID:910
Depositado Por:Marisa G. Velazco Aldao
Depositado En:28 Abr 2021 09:39
Última Modificación:28 Abr 2021 09:45

Personal del repositorio solamente: página de control del documento