IEEE Geoscience and Remote Sensing Magazine - December 2021 - 103

[2]. This type of imagery enables the simultaneous registration
of multiple parts of the electromagnetic spectrum,
providing invaluable information for many detection, classification,
and unmixing problems [3]. As a result, today,
remote sensing HSI is used in many commercial, scientific,
and defense areas, including precision agriculture, mining,
forestry, coastal and oceanic observation, intelligence, and
disaster monitoring [3]-[6]. Due to the growing quantity
of deployed sensors [7], the number of public and private
remote sensing stakeholders [4], and the ongoing effort to
improve the analysis of retrieved images [8]-[19], the importance
of HSI is likely to increase in the future.
Images produced by multispectral and hyperspectral
sensors consist of multiple spectral bands, instead of the
three-red, green, and blue-present in traditional color
images. Depending on the application and the available
hardware, the number of registered bands can be on the order
of tens, hundreds, and even thousands [20]. Thus, HSI
generates significantly larger volumes of data compared
to traditional imagers. Moreover, the spatial resolution of
the deployed sensors also follows a rising trend, further
increasing the amount of data produced. For instance, the
HyspIRI sensor developed by NASA can produce up to 5 TB
of data per day [21]. However, the downlink channel capacity
between the remote sensing devices and the ground stations
is constrained, which limits the amount and quality
of the retrieved data [22].
Data compression is typically applied to reduce the
amount of data to be downloaded, hence improving effective
transmission capacity [23]-[27]. Due to hardware and
energy constraints, employed algorithms must be tailored
to attain a beneficial tradeoff between complexity and efficiency
[22], [28]. When lossless compression is applied
to images, the resulting compressed data suffice to reconstruct
identical copies of the originals. On the other hand,
lossy compression enables the transmission of even smaller
data volumes at the cost of the reconstructed images not
being identical to the originals. Among lossy compression
algorithms, those that provide user-controlled bounds on
the maximum error introduced in any sample are referred
to as near lossless.
In spite of the distortion introduced by lossy and nearlossless
methods, several studies have concluded that reconstructed
images can be successfully used for the intended
analysis tasks [29]. This is sometimes observed for
compressed images up to 25-times smaller than the original
ones [30]. Notwithstanding, a successful analysis can
be performed only when the amount of loss is adequate
for the type of images and the task at hand [29], [31]. One
of the main advantages of near-lossless compressors is
that they guarantee the accuracy of all the reconstructed
samples in an image. This is in contrast to regular lossy
compression approaches, which typically provide competitive
average distortion results but no assurance about
the fidelity of any given set of samples. Regardless of the
employed compression regime, compression algorithms
DECEMBER 2021 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
must meet very stringent limitations in terms of complexity
and required computational resources [32]. This
constraint is particularly relevant for small satellites and
CubeSats, which have attracted much scientific and industrial
interest recently [4], [33].
The CCSDS, founded in 1982, publishes the standards
for spaceflight communication used in more than 900
space missions to date. (An updated list of space missions
using CCSDS standards can be found at https://public.
ccsds.org/implementations/missions.aspx.) CCSDS standards
enable cooperation among space agencies and with
industrial associates, seeking enhanced interoperability,
reliability, and cost-effectiveness. The latest CCSDS compression
standard (CCSDS 123.0-B-2, " Low-Complexity
Lossless and Near-Lossless Multispectral and Hyperspectral
Image Compression [34]), the central topic of this
article, supersedes CCSDS 123.0-B-1 [35] while maintaining
backward compatibility. In the CCSDS naming
convention, suffixes " -1 " and " -2 " denote the first and second
issues, respectively, of a standard. Hereafter, CCSDS
123.0-B-1 and CCSDS 123.0-B-2 are also denoted as Issue
1 and Issue 2, respectively.
Perhaps the most relevant novel feature of Issue 2 is
a new near-lossless compression regime, enabled by a
closed-loop scalar quantizer in the prediction stage [36].
Note that this in-loop quantization approach enables a
higher compression performance than quantization of
input samples before prediction [26]. With this new feature,
users can specify the maximum error limits-absolute
and/or relative-introduced in the decompressed
images. Fidelity settings can vary from band to band and
can be periodically updated within an image. Another
new feature of Issue 2 is a hybrid entropy coder option. It
is specifically designed to provide improved performance
on low-entropy data, i.e., for the case when prediction errors
tend to be small compared to the quantizer stepsize.
The hybrid encoder extends the sample-adaptive codes of
CCSDS 123.0-B-1 with 16 additional variable-to-variablelength
codes, which can represent multiple input symbols
using a single codeword. To guarantee backward compatibility,
both lossless and near-lossless compression can be
performed with either of CCSDS 123.0-B-1's original entropy
coders or with the new hybrid option. A third novelty
in the new standard is a new mode within the predictor
stage called narrow local sums, which are designed
to facilitate the design of efficient hardware implementations.
Yet another change introduced in the new standard
is added support for optional supplementary information
tables, which can provide ancillary image or instrument
information, e.g., to identify the wavelengths associated
with each spectral band.
This article provides a comprehensive overview of Issue 2,
paying special attention to the new concepts and capabilities
not present in Issue 1. The content hereafter presented
extends those presented in a previous conference work [36].
The following overview is more in depth, it assumes no
103
https://public.ccsds.org/implementations/missions.aspx https://public.ccsds.org/implementations/missions.aspx
IEEE Geoscience and Remote Sensing Magazine - December 2021

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