CONVOLUTIONAL NEURAL NETWORKS FOR DETECTION AND CLASSIFICATION OF MARITIME VESSELS IN ELECTRO-OPTICAL SATELLITE IMAGERY
Loading...
Authors
Rice, Katherine
Subjects
deep learning
machine learning
image classification
object detection
transfer learning
satellite imagery
single shot detector
convolutional neural network
machine learning
image classification
object detection
transfer learning
satellite imagery
single shot detector
convolutional neural network
Advisors
Whitaker, Lyn R.
McCarrin, Michael R.
Date of Issue
2018-12
Date
Publisher
Monterey, CA; Naval Postgraduate School
Language
Abstract
The ability to locate and identify vessels of interest in satellite imagery plays a vital role in maintaining maritime security. Recent studies have demonstrated that convolutional neural networks can be used to automatically classify or detect ships in satellite images; however, this technique requires large amounts of training data and computational power that may not be readily available in an operational environment. We seek to show that the process of transfer learning can be used to adapt open source convolutional neural network architectures pre-trained on large datasets to Department of Defense-specific image classification and detection tasks. We test this hypothesis by retraining both the VGG-16 image classification architecture and a VGG-16 based Single Shot Detector on a dataset comprised of satellite images containing ships. We first examine the performance of these retrained networks on the single category task of classifying or detecting ships in satellite imagery. We then evaluate model performance on datasets in which a fraction of the images contains blur and noise to simulate degraded satellite imagery. Finally, we test the models’ ability to distinguish between subcategories of ships. We show that transfer learning can be leveraged to reduce both the size of the training set and the training time required to produce an effective classification or detection model to meet the Department of Defense’s 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.