IEEE Geoscience and Remote Sensing Magazine - June 2016 - 88

acKnoWleDgMent
This work is supported by a grant from Singapore's Defense
Science and Technology Agency. B. Wen and S. Winkler
are also supported by the research grant for the HumanCentered Cyber-Physical Systems Program at the Advanced
Digital Sciences Center from Singapore's Agency for Science, Technology, and Research, A*STAR.

(a)

(b)

figure 11. The real large-scale night-time cloud/sky image
denoising result, with a regional zoom-in for comparison: (a) a real
noisy cloud image and (b) a denoised image obtained using an
online transform learning-based denoising scheme [88].

K-SVD since they are using either a more complicated model or supervised learning. The denoising result using the
GMM-based method is shown in Figure 9(b).
WSIs continuously generate large-scale cloud image data
that need to be processed efficiently. Although learning-based
algorithms can provide a promising performance in applications such as denoising, most of them are batch algorithms.
Consequently, the storage requirements of batch methods such
as K-SVD and OCTOBOS increase with the size of the data set;
furthermore, processing real-time data in batch mode translates
to latency. Thus, online versions of learning-based methods
[86], [87] are needed to process high-resolution WSI data. These
online learning schemes are more scalable to big-data problems
by taking advantage of stochastic learning techniques.
Here, we show an example of denoising a color image
measurement of 3,000 # 3,000 pixels that is generated by
WAHRSIS at night using online transform learning [88].
The denoising results are illustrated in Figure 11. It is important to note that such a method is also capable of processing
real-time high-dimensional data [89]. Thus, it can easily be
extended to applications involving multitemporal satellite
images and multispectral data in remote sensing.
concluSion
In this article, we have provided an overview of recent developments in machine learning for remote sensing, using
examples from ground-based image analysis. Sensing the
earth's atmosphere using high-resolution ground-based
sky cameras provides a cheaper, faster, and more localized
manner of data acquisition. Due to the inherent highdimensionality of the data, it is expensive to directly use
raw data for analysis. We have introduced several featureextraction techniques and demonstrated their properties
using illustrative examples. We have also provided extensive experimental results in segmentation, classification,
and denoising of sky/cloud images. Several techniques
from machine learning and computer vision communities
have been adapted to the field of remote sensing and often
outperform conventional heuristic approaches.
90

author inforMation
Soumyabrata Dev (soumyabr001@e.ntu.edu.sg) graduated summa cum laude from the National Institute of
Technology Silchar, India, with a B.Tech. in electronics and
communication engineering in 2010. From 2010 to 2012,
he worked with Ericsson as a network engineer. Currently,
he is pursuing a Ph.D. degree in the School of Electrical and
Electronic Engineering, Nanyang Technological University,
Singapore. From August to December 2015, he was a visiting student at the Audiovisual Communication Laboratory,
École Polytechnique Fédérale de Lausanne, Switzerland.
His research interests include remote sensing, statistical
image processing, and machine learning. He is a Student
Member of the IEEE.
Bihan Wen (bwen3@illinois.edu) received a B.Eng. degree in electrical and electronic engineering from Nanyang
Technological University, Singapore, in 2012 and an M.S.
degree in electrical and computer engineering from the
University of Illinois at Urbana-Champaign in 2015. He
is currently pursuing a Ph.D. degree at the University of
Illinois at Urbana-Champaign. His current research interests include signal and image processing, machine learning, sparse representation, and big data applications. He is a
Student Member of the IEEE.
Yee hui lee (EYHLee@ntu.edu.sg) received the B.Eng.
(honors) and M.Eng. degrees from the School of Electrical and Electronics Engineering at Nanyang Technological
University, Singapore, in 1996 and 1998, respectively, and a
Ph.D. degree from the University of York, United Kingdom,
in 2002. Currently, she is an associate professor and assistant chair (students) at the School of Electrical and Electronic
Engineering, Nanyang Technological University, where she
has been a faculty member since 2002. Her research interests
include channel characterization, rain propagation, antenna
design, electromagnetic bandgap structures, and evolutionary techniques. She is a Senior Member of the IEEE.
Stefan Winkler (stefan.winkler@adsc.com.sg) received B.Eng. and M.Eng. degrees from the University of
Technology Vienna, Austria, and the Ph.D. degree from the
École Polytechnique Fédérale de Lausanne, Switzerland,
in 1996 and 2001, respecitively. He cofounded Genista,
worked in several large corporations in Europe and the United States, and held faculty positions at the National University of Singapore and the University of Lausanne, Switzerland. He is currently a principal scientist and director of the
Video and Analytics Program at the University of Illinois'
Advanced Digital Sciences Center in Singapore. He has published over 100 papers and the book Digital Video Quality.
ieee Geoscience and remote sensing magazine

june 2016

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - June 2016