Abstract

Many tasks in the field of computer vision rely on an underlying change detection algorithm in images or video sequences. Although much research has focused on change detection in consumer images, there is little work related to change detection on aerial imagery, where individual images are recorded from aerial platforms over time.

This thesis presents two deep learning approaches for detection in aerial images. Both systems leverage Spatial Transformer Networks (STN) that identify the coordinate transformation for their localization capabilities. The first approach is based on a semisupervised approach which learns to locate changes within a difference image. The second is a fully-supervised approach which learns to locate and discriminate relevant targets. The supervised approach is shown to locate nearly 78% of positive samples with an Intersection Over Union (IOU) criterion of over 0.5, and nearly 94% of positive samples with an IOU over 0.3.

Library of Congress Subject Headings

Remote-sensing images--Data processing; Optical pattern recognition; Computer vision; Image processing--Digital techniques

Publication Date

11-2016

Document Type

Thesis

Student Type

Graduate

Degree Name

Computer Engineering (MS)

Department, Program, or Center

Computer Engineering (KGCOE)

Advisor

Andreas Savakis

Advisor/Committee Member

John Kerekes

Advisor/Committee Member

Dhireesha Kudithipudi

Comments

Physical copy available from RIT's Wallace Library at G70.4 .C44 2016

Campus

RIT – Main Campus

Plan Codes

CMPE-MS

Download

Share

COinS