Publication: Clasificador completo de células sanguíneas mediante una FPGA de bajo coste, MATLAB y SIMULINK
Loading...
Official URL
Full text at PDC
Publication Date
2017
Authors
Advisors (or tutors)
Barrio García, Alberto del
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citations
Exportar
Abstract
Actualmente la creación de algoritmos para FPGA tiene un auge considerable gracias a las prestaciones que presentan estos chips para el procesamiento de información. Para mejorar la experiencia en el desarrollo de hardware compañías como MathWorks ofrecen herramientas de síntesis de alto nivel para acelerar el proceso de diseño. En el presente trabajo se analizó el uso de Matlab y Simulink, así como toolbox, enfocados al desarrollo de hardware. Para este análisis se crearon dos casos de estudio para el reconocimiento de células en imágenes de microscopía utilizando redes neuronales. Para cada caso se implementaron modelos basados en el algoritmo DCT, pero empleando diferentes estrategias de extracción de características.
Una vez generados los modelos se utilizó el entorno de co-simulación FPGA in-the-loop ofrecido por Simulink para ejecutar los modelos directamente en la FPGA. Se analizaron y presentaron los resultados obtenidos por precisión y por tiempos de ejecución. Adicionalmente se realizan comparaciones entre los modelos generados y su contraparte en software con el fin de validar el grado de mejora obtenido en los tiempos de ejecución con uso de hardware para la aceleración de cómputo.
Developing algorithms for hardware, especially for FPGAs, is gaining popularity nowadays. This is thanks to the performance provided by these chips in the different areas. To improve the experience of hardware development, companies as MathWorks offer high-level synthesis tools to speed up the design process. In this work, we analyzed the use of Matlab and Simulink, as well as the hardware development toolbox. For this analysis, two different study cases were created. Both studies focused on the recognition of cells in microscopy images using neural networks. For each case, a model based on the DCT algorithm was implemented, but using different strategies for the feature extraction. Once the models were generated, we used the FPGA in the loop environment offered by Simulink to run the models directly into the FPGA. The results obtained in terms of precision and the execution times were analyzed and presented. In addition, several comparisons were made between the generated models and their equivalent in software to validate the degree of achieved improvement.
Developing algorithms for hardware, especially for FPGAs, is gaining popularity nowadays. This is thanks to the performance provided by these chips in the different areas. To improve the experience of hardware development, companies as MathWorks offer high-level synthesis tools to speed up the design process. In this work, we analyzed the use of Matlab and Simulink, as well as the hardware development toolbox. For this analysis, two different study cases were created. Both studies focused on the recognition of cells in microscopy images using neural networks. For each case, a model based on the DCT algorithm was implemented, but using different strategies for the feature extraction. Once the models were generated, we used the FPGA in the loop environment offered by Simulink to run the models directly into the FPGA. The results obtained in terms of precision and the execution times were analyzed and presented. In addition, several comparisons were made between the generated models and their equivalent in software to validate the degree of achieved improvement.
Description
Máster en Ingeniería Informática, Facultad de Informática, Departamento de Arquitectura de Computadoras y Automática, curso 2016-2017