Deep learning for media analysis in defense scenarios--an evaluation of an open-source framework for object detection in intelligence-related image sets
Loading...
Authors
Paul, Taylor H.
Subjects
deep learning
machine learning
object detection
computer forensics
open source
TensorFlow
machine learning
object detection
computer forensics
open source
TensorFlow
Advisors
Kölsch, Mathias
Date of Issue
2017-06
Date
Jun-17
Publisher
Monterey, California: Naval Postgraduate School
Language
Abstract
The Department of Defense struggles to develop and maintain cutting-edge software through the Defense Acquisition System. The pace of improvements in machine learning algorithms and software suggests the organization will fail to rapidly develop systems incorporating the latest innovations to meet its intelligence-related media analysis needs. In contrast, the trend of industry and academia releasing algorithms and software under permissive licenses bestows defense organizations with an opportunity. These groups can potentially overcome resource shortfalls and long acquisition timelines by implementing machine learning solutions with open-source software.We test this hypothesis by employing an open-source software library to evaluate publicly available deep learning algorithms on three prior defense-related datasets. We then compare performance of deep convolutional neural networks to past methods for detecting AK-47s, ships, and screenshots in images. Applying proven algorithms through the software framework, we test three object detectors that exceed or match classification performance for all three experiments in a third of the development time available to designers of the previous algorithms. We relate these tests to defense scenarios in order to provide a logical argument and empirical measure of the utility of open-source machine learning frameworks to meet the Department of Defense's intelligence-related media analysis needs.
Type
Thesis
Description
Series/Report No
Department
Computer Science (CS)
Organization
Identifiers
NPS Report Number
Sponsors
Funder
Format
Citation
Distribution Statement
Approved for public release; distribution is unlimited.
Rights
This publication is a work of the U.S. Government as defined in Title 17, United States Code, Section 101. Copyright protection is not available for this work in the United States.