IEEE Geoscience and Remote Sensing Magazine - March 2017 - 8

Advanced
Spectral
Classifiers for
Hyperspectral
Images
A review
PEDRAM GHAMISI, JAVIER PLAZA,
YUSHI CHEN, JUN LI, AND ANTONIO PLAZA

yperspectral image classification has been a vibrant
area of research in recent years. Given a set of observations, i.e., pixel vectors in a hyperspectral image, classification approaches try to allocate a unique label to each pixel
vector. However, the classification of hyperspectral images
is a challenging task for a number of reasons, such as the
presence of redundant features, the imbalance among the
limited number of available training samples, and the high
dimensionality of the data.
The aforementioned issues (among others) make the
commonly used classification methods designed for the
analysis of gray scale, color, or multispectral images inappropriate for hyperspectral images. To this end, several
spectral classifiers have been specifically developed for
hyperspectral images or carried out on such data. Among
those approaches, support vector machines (SVMs), random forests (RFs), neural networks, deep approaches,
and logistic regression-based techniques have attracted
great interest in the hyperspectral community. This article
reviews most of the existing spectral classification approaches in the literature. It also critically compares the
most powerful hyperspectral classification approaches
from different points of view, including their classification accuracy and computational complexity. The article
goes on to provide several hints for readers about the logical choice of an appropriate classifier based on the application at hand.
Digital Object Identifier 10.1109/MGRS.2016.2616418
Date of publication: 16 March 2017

CLASSIFYING HYPERSPECTRAL DATA
Imaging spectroscopy (also known as hyperspectral imaging)
is an important technique in remote sensing (RS). Hyperspectral imaging sensors often capture data from the visible
through the near-infrared wavelength ranges, thus providing
hundreds of narrow spectral channels from the same area on
the surface of the earth. These instruments collect data consisting of a set of pixels represented as vectors, in which each
element is a measurement corresponding to a specific wavelength. The size of each vector is equal to the number of spectral channels or bands. Hyperspectral images usually consist
of several hundred spectral data channels for the same area on
the earth's surface; while, in multispectral data, the number
of spectral channels is usually up to tens of bands [1]. The detailed spectral information collected by hyperspectral sensors
increases the capability of discriminating between different
land-cover classes with increased accuracy. Several operational hyperspectral imaging systems are currently available, providing a large volume of image data that can be used for a wide
variety of applications, such as in ecology, geology, hydrology,
precision agriculture, and military applications.
Due to the detailed spectral information available from
the hundreds of narrow bands collected by hyperspectral
sensors, the accurate discrimination of different materials
is possible. This fact makes hyperspectral data a valuable
ieee Geoscience and remote sensing magazine

march 2017

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2017