-
Notifications
You must be signed in to change notification settings - Fork 0
/
signature_ai.tex
186 lines (147 loc) · 6.72 KB
/
signature_ai.tex
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
\documentclass[pdftex,letterpaper,11pt]{article}%
\usepackage[export]{adjustbox}
\usepackage{multirow}
\usepackage[arabicsections]{dpugatex}
%\usepackage{dpppl}
%\usepackage{lscape}
\usepackage{pdflscape}
\usepackage{longtable}
\usepackage{tikz}
\usepackage[british]{babel}
\usepackage{amssymb,amsmath,warpcol,url,ctable,multirow,caption,threeparttable,float,soul,gensymb}
%\usepackage[noend]{algpseudocode}
\makeatletter
\def\BState{\State\hskip-\ALG@thistlm}
\makeatother
\usepackage{subcaption}
\usepackage{xcolor,colortbl}%Per colorejar cel·les de les taules
\definecolor{verylightgray}{gray}{0.90}
\newcommand{\verylightgray}[1]{\cellcolor{verylightgray}#1}
\usepackage{arydshln} %Per dashed lines
\setlength\dashlinedash{1.5pt}
\setlength\dashlinegap{2.5pt}
\setlength\arrayrulewidth{0.8pt}
\usepackage[utf8]{inputenc}
\usepackage{eurosym}
\usepackage{soul}
\setstcolor{red} % Tatxat vermell pel text a eliminar!!!!
\definecolor{lightgreen}{rgb}{0.564706,0.933333,0.564706}
\newcommand{\miquel}[1]{{\sethlcolor{lightgreen} \hl{#1}} } % To highlight with light green Miquel's comments.
\soulregister\citet7 % These \soulregister commands allow us to avoid problems between soul functions (\hl, \st, ...) and these other functions (\citet,\citep,\ref,\footnote, ...). Include more if it is necessary.
\soulregister\citep7
\soulregister\cite7
\soulregister\ref7
\soulregister\footnote7
\soulregister\textit7
\soulregister\textbf7
\soulregister\textsc7
\soulregister\`7 % for open accents.
\soulregister\'7 % for closed accents.
\soulregister\% 7 % for \%age symbol
\usepackage{tabularx,tabulary}
\usepackage[osf]{mathpazo} % Amb aquest paquet faig ús de la lletra Palatino, la que fan servir Puga, Duranton et al. Amb l'opció [osf] la numeració és "old fashion", però és incompatible amb un títol en negreta i small caps (desapareixen les small caps!). Per tant, per ara opto per NO aplicar aquesta opció.
\usepackage[abs]{overpic}
\usepackage{hyperref}
\definecolor{darkblue}{rgb}{0.0,0.0,0.3}
\hypersetup{colorlinks=true, breaklinks=true, citecolor= darkblue, linkcolor=blue, urlcolor=blue}
\usepackage{chngcntr}
\newcommand{\mc}[3]{\multicolumn{#1}{#2}{#3}}
\newcommand{\mcc}[1]{\multicolumn{1}{c}{#1}}
\newcommand{\mcb}[1]{\multicolumn{1}{c}{\bf #1}}
\newcommand{\mr}[3]{\multirow{#1}{#2}{#3}} % Per fer Files múltiples (seguint el punt de vista de les columnes múltiples.
\newcommand{\mg}[3]{\multicolumn{#1}{#2}{\verylightgray #3}} %Combino el color de les cel·les amb al command de les columnes múltiples.
%\hypersetup{%
% pdftitle={Race and neighborhoods in the 21$^{\text{st}}$ century},%
% pdfauthor={Jorge De la Roca (NYU) , Ingrid Gould Ellen (NYU) and Katherine O'Regan (NYU)},%
% pdfkeywords={race segregation, discrimination}}
%\pdfOpenFitWidth
%\pdfShowBookmarks
\onehalfspacing%
%\doublespacing%
\newenvironment{tablenote}[1]{\begin{list}{}{\vskip-5mm\relax
\setlength{\leftmargin}{#1} \setlength{\rightmargin}{\leftmargin}}
\item[]\footnotesize\vskip-7pt
{\em Notes}:\space\ignorespaces}{\end{list}}
\newcommand{\jdlradded}[1]{#1}
\newcommand{\dpadded}[1]{#1}
\newcommand{\jdlrdeleted}[1]{}
\newcommand{\dpdeleted}[1]{}
\newcommand{\jdlrcomment}[1]{}
\newcommand{\dpcomment}[1]{}
\newcommand{\comment}[1]{}
\newcommand{\martin}[1]{{\color{blue} Martin: [{#1}]}}
\newcommand{\dani}[1]{{\color{purple} Dani: [{#1}]}}
\begin{document}
\begin{titlepage}
\vspace*{1ex}
\begin{minipage}{\textwidth}
\begin{center}%
{\textsb{\LARGE \textit{``Decoding (urban) form and function using spatially explicit deep learning''}}}\\[4ex]%
{\Large\textsb{Martin Fleischmann}\footnote[1]{Department of Social Geography and Regional Development, Charles University, Albertov 2038/6, Praha 2 128 00, Czechia}\footnote[2]{Geographic
Data Science Lab, Department of Geography and Planning, University of
Liverpool, Roxby Building, 74 Bedford St S, Liverpool, L69 7ZT, United
Kingdom}\footnote[4]{
E-mail: \url{[email protected]}; website: \url{https://martinfleischmann.net/}.}}\\[1mm]
%
{\Large\textbf{Daniel Arribas-Bel}\footnotemark[2]\footnote[3]{E-mail: \url{[email protected]}; phone: +44 (0)151 795 9727; website: \url{http://darribas.org}.
}}\footnote[5]{The Alan Turing Institute, British Library, 96 Euston Road, London, England, NW1 2DB, United Kingdom}
\\[1mm]
{\large\textit{Department of Social Geography and Regional Development, Charles University} }\\
{\large\textit{Geographic Data Science Lab, University of Liverpool} }\\
{\large\textit{The Alan Turing Institute} }\\[2.5ex]
%
\date{\today}\vspace{1.5ex}
%
\end{center}
%
\begin{abstract}
This paper explores how can geographical dimension be incorporated into deep
learning designed to understand the composition of urban landscapes based on
Sentinel 2 satellite imagery. Compared to standard computer vision, satellite
imagery is unique as images sampled from the data form a continuous array, rather
than being fully independent. We argue that the spatial configuration of the images
is as important as the content of each image when attempting to capture a pattern
that reflects the structure of the urban environment. We propose a series of
approaches explicitly incorporating spatial dimension in the predictive pipeline
based on the EfficientNetB4 convolutional neural network (CNN) and experimentally
test their effect on model performance. The experiments in this study cover the
scale of the sampled area, the effect of spatial augmentation, and the role of
modelling (logit ensemble and histogram-based gradient-boosted classifiers) with and
without the spatial context on the outputs of the neural network-generated vector of
probabilities while trying to predict spatial signatures, a classification of
primarily urban landscape based on form and function. The results suggest that
certain ways of embedding spatial information, especially in the modelling step,
consistently significantly improve the prediction accuracy and shall be considered
on top of standard CNNs.
\end{abstract}
%
\vspace{1.5ex}
%
Key words: \hskip.25em spatial signatures, classification, remote sensing, artificial intelligence, open data\\
%
\vspace*{-1.5ex}
%
\end{minipage}
\end{titlepage}
% Section 1 - Intro
\input{s1_intro.tex}
% 1500 words
% Section 2 - Materials & Methods
\input{s2_materials_methods.tex}
% 3000 words
% Section 3 - Results
\input{s3_results.tex}
% 500 words
% Section 4 - Discussion
\input{s4_discussion.tex}
% 1000 words
\section*{Data and code availability statement}
All the data and code will be available on a public repository with DOI upon acceptance
of the article to ensure the anonymity of a double-blind review.
\clearpage
\bibliographystyle{apalike}
\bibliography{refs}
\clearpage
\input{sa_appendix.tex}
\clearpage
\end{document}