IEEE Geoscience and Remote Sensing Magazine - March 2021 - 28

and MSDCNN methods show overall satisfactory performance. Among them, the MTF-GLP and MTF-GLP-HPMPP methods show good performance in the reduced-resolution experiment and relatively poorer performance in the
full-resolution experiment. Conversely, the MTF-GLP-HPM
and MTF-GLP-CBD methods show good performance in
the full-resolution experiment and poor performance in
the reduced-resolution experiment. The BDSD and MSDCNN methods show excellent and robust performance in
the reduced- and full-resolution experiments.
For the experiment results based on GF-1 images, the
BDSD, ATWT-model 2 (M2), ATWT-model 3 (M3), MTFGLP-HPM, MTF-GLP-CBD, and MSDCNN methods show
overall satisfactory performance. While the best performance is shown for the MTF-GLP-CBD method in the
full-resolution experiment, the approach had a relatively

poorer performance in the reduced-resolution experiment,
especially in the ERGAS and SAM quantitative-evaluation
indices. The DL-based MSDCNN method shows the best
performance and has competitive advantages over the
other pansharpening methods in the reduced-resolution
experiment. In addition, it shows robust performance in
the full-resolution experiment.
In the WorldView-4 experiment, the BDSD, PRACS, MTFGLP, MTF-GLP-HPM, MTF-GLP-CBD, and MSDCNN methods show overall satisfactory performance. Among them,
the MSDCNN method has excellent experiment results in
the reduced- and full resolution experiments, especially in
the quantitative-evaluation indices of the ERGAS, SAM, D s,
and QNR. The MTF-GLP-HPM method shows good performance in the full-resolution experiment and relatively
poorer performance in the reduced-resolution experiment.

TABLE 3. THE PANSHARPENING METHODS IN THE EXPERIMENTS.

28

METHOD

BASIC MEANING AND PARAMETER SETTINGS

Generalized IHS [3], [19]

A generalization of the traditional IHS pansharpening method [17] for the fusion of more than three MS
bands; it is also called the fast generalized IHS fusion method

Brovey [3], [102]

A CS-based method under the general understanding; it has been commercialized in professional software, such as ENVI

PCA [3], [17]

A typical CS-based method founded on the PCA transformation; it has been commercialized in professional software, such as ENVI

BDSD [3], [61]

A data-dependent, self-adaptive CS-based approach featuring two advantages: 1) the optimal solution of
the spatial-detail extraction and injection based on the minimum mean square error and 2) parameter
estimation based on local regions

GS [3], [39]

A typical CS-based method built on the GS transformation; it has been commercialized in professional
software, such as ENVI

GSA [3], [38]

An improved GS method in terms of the adaptive calculation of the spectral band combination weight
based on a linear-regression algorithm

PRACS [3], [62]

A CS-based method founded on the partial replacement of the intensity component; the Pan is simulated
based on the weighted average of the Pan and MS bands instead of directly using the original Pan to
replace the intensity component

HPF [3], [17]

A typical MRA-based method; in the experiments, the default parameter with a 5 × 5 box filter was used

Smoothing filter-based intensity
modulation (SFIM) [3], [103], [104]

An enhanced HPF method based on the spatial-detail injection scheme of HPM, which is calculated based
on the ratio of the resampled MS image and the low-pass filtering Pan image

Indusion [3], [105]

Indusion combines induction and fusion; it is a fusion method based on the induction-scaling technique
and profits from multiple equalization steps to improve performance

MTF-GLP [3], [41]

A popular MRA-based method founded on the GLP with an MTF filter; the unitary detail-injection
model is applied

MTF-GLP-HPM [3], [106]

The MTF-GLP with the spatial-detail injection scheme of the HPM

MTF-GLP-HPM-PP [3], [107]

The MTF-GLP-HPM followed by a PP phase [107] to correct the noise at the edge regions

MTF-GLP-CBD [3], [50]

The MTF-GLP with the spatial-detail injection scheme of the CBD, which is locally optimized by patching
the image in nonoverlapping zones

ATWT [3], [15], [67]

An MRA-based method built on the ATWT with a unitary injection model

ATWT-M2 [3], [65]

The ATWT with model 2 proposed in [65], which is based on moment matching to adjust the wavelet coefficients of the Pan image

ATWT-M3 [3], [65]

The ATWT with model 3 proposed in [65], which is based on least-square fitting to adjust the wavelet coefficients of the Pan image

AWLP [3], [40]

A generalization of the AWL pansharpening method [15] in terms of a proportional spatial-detail injection
weight, relying on the original MS-band radiance proportionality

MSDCNN [45]

A DL-based method founded on an MSDCNN

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

MARCH 2021
IEEE Geoscience and Remote Sensing Magazine - March 2021

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