Accession Number : AD1023022


Title :   Convolutional Neural Network on Embedded Linux(trademark) System-on-Chip: A Methodology and Performance Benchmark


Descriptive Note : Technical Report


Corporate Author : Space and Naval Warfare Systems Center Pacific San Diego United States


Personal Author(s) : Gebhardt,Daniel ; parikh,Keyur ; Dzieciuch,Iryna


Full Text : https://apps.dtic.mil/dtic/tr/fulltext/u2/1023022.pdf


Report Date : 01 May 2016


Pagination or Media Count : 18


Abstract : Deep convolutional neural networks (CNNs) detect and classify features of interest in sensory input data. There is a need to investigate how best to implement CNNs for Navy and Department of Defense (DoD) use in platforms with minimal size, weight, and power (SWaP) capacity, since much academic research focuses solely on achieving the highest performance on a specific dataset with minimal concern of compute resources. This report describes a methodology, configuration, and experimental results of a first step in this study a baseline for comparison of benchmarking metrics. A baseline is important for quantifying any further results and to estimate potential benefits of new and more advanced ideas.


Descriptors :   field programmable gate arrays , embedded systems , computing system architectures , machine learning , network architecture , signal processing , systems engineering , artificial neural networks , operating systems , MULTISENSORS , CLASSIFICATION , FEATURE EXTRACTION


Subject Categories : Computer Systems
      Computer Hardware


Distribution Statement : APPROVED FOR PUBLIC RELEASE