IEEE Geoscience and Remote Sensing Magazine - September 2021 - 43

Classification of
Remote Sensing Data
With Morphological
Attribute Profiles
©SHUTTERSTOCK.COM/LAVIZZARA
A decade of advances
DEISE SANTANA MAIA, MINH-TAN PHAM,
ERCHAN APTOULA, FLORENT GUIOTTE,
AND SÉBASTIEN LEFÈVRE
AP variations is provided as well, not only in terms of classification
performance but, for the first time, in terms of
their generalization capacity, too.
M
orphological attribute profiles (APs) are among the
most prominent methods for spatial-spectral pixel
analysis of remote sensing images. Since their introduction
a decade ago to tackle land cover classification, many studies
have been contributed to the state of the art, focusing
not only on their application to a wider range of tasks but
also on their performance improvement and extension to
more complex Earth observation data.
Despite the overwhelming proliferation of deep learning-based
methods in the past five years, APs are far from
obsolete, due mainly to their high flexibility, low computational
cost, lower training data requirement, and rigorous
mathematical foundation. In this survey, an entire decade's
worth of more than 100 AP-related contributions to the
field of remote sensing have been compiled, providing an
extensive panorama of this robust and effective tool. Moreover,
a collective experimental comparison of the reviewed
Digital Object Identifier 10.1109/MGRS.2021.3051859
Date of current version: 1 March 2021
SEPTEMBER 2021 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
BACKGROUND
Classification constitutes one of the paramount tasks of
remote sensing image analysis for Earth observation. Its
performance is critical for the success of land use and cover
mapping and monitoring. The rapid development of highresolution
(HR) and very-HR (VHR) image acquisition
technologies has led to increasingly complex images with
higher levels of detail. Consequently, the importance of the
joint exploitation of spatially contextual information along
with spectral pixel characteristics became clear early on [1].
Morphological profiles (MPs) [2] were introduced almost
two decades ago to address this exact issue. In essence, they
produce multiscale descriptions of their input through the
application of a sequence of morphological reconstructionbased
filters using structuring elements of various sizes (and
shapes) [2]. Their rigorous mathematical foundation and inherent
ability to capture spatial-spectral information have
led to the development of several variants [3]-[6].
However, as spatial resolutions and image sizes progressively
increased, their relatively high computational cost
has been drastically accentuated. Moreover, their initially
celebrated capacity of capturing size and shape variations
eventually became extremely limited when confronted
with thematic classes differing in terms of alternative
properties, such as contrast, homogeneity, and so on. In
an effort to overcome these shortcomings, MPs have been
0274-6638/21©2021IEEE
43
http://www.SHUTTERSTOCK.COM/LAVIZZARA
IEEE Geoscience and Remote Sensing Magazine - September 2021

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - September 2021
