Results and Discussions - Hyperspectral Image Fusion - page 149

Table 10.1 Description of the hyperspectral datasets used in this monograph

Data name Sensor Dimensions Geographical location Region depicted

37 ◦ 44 N 121 ◦ 80 W Moffett 2 AVIRIS 614 × 512 × 224 Moffett Field, CA

37 ◦ 44 N 122 ◦ 21 W

Moffett 3 AVIRIS 614 × 512 × 224 Moffett Field, CA

38 ◦ 44 N 115 ◦ 78 W Lunar 2 AVIRIS 614 × 512 × 224 Lunar Lake, NV

37 ◦ 47 N 117 ◦ 49 W

Geological Hyperion 512 × 256 × 242 West-central Nevada,

20 ◦ 24 S 164 ◦ 30 E Coral Hyperion 512 × 256 × 242 Coral Sea, New Caledonia

37 ◦ 47 N 122 ◦ 13 W alo Urban Hyperion 512 × 256 × 242 lto, A

We choose the remaining 3 datasets from the Hyperion imaging sensor used in
the EO-1 spacecraft for earth observation. 2 These data consist of 242 bands covering
the bandwidth from 0.4 to 2 m. The data processed for the terrain correction are

designated as the Level G1. The dimensions of bands in the original are

. 5 μ

( 2905 ×

256 pixels. However, for faster computation, we split the data across their length to

obtain several datasets of a smaller dimensions

)

pixels. We select datasets
depicting a wide variety of features. The first data provide some of the geological
features of west-central Nevada area, and hence it will be called as the geological
data. We have already presented this dataset for the consistency analysis of a fusion
scheme in Chap. 9 . We also select a hyperspectral dataset capturing some regions
of the coral reefs near Australia. This dataset has been referred to as the coral data.
Since we have discussed four different techniques for fusion of hyperspectral data
in Chaps. 3 , 5 - 7 of this monograph, we illustrate the corresponding results of fusion
finally using the urban dataset for every solution. This urban dataset is obtained by
the Hyperion, which captures some of the urban regions of Palo Alto, CA, and hence
the name. Details of these datasets can be found in Table 10.1 .

( 512 × 256 )

10.3 Results of Fusion

Let us begin with the performance evaluation of the moffett 2 data. The results of
fusion using the techniques described in previous chapters are shown in Fig. 10.1 .
The bilateral filtering-based technique provides a sharp image of the scene as shown
in Fig. 10.1 a. The moffett 2 scene contains a large number of smaller objects mainly
on the top and the right sides of the image, which are well-discriminable in this
image. The output of the Bayesian fusion technique from Fig. 10.1 b, however, appears
sharper than the first image. Although this image has a slightly brighter appearance,
the sharpness in the image appears to be independent of the higher mean value.
This sharpness is due to the fact that the pixels with higher gradient values have
contributed more towards the resultant final image. The appearance of the output of

2

Hyperion data available from the U.S. Geological Survey: http://eo1.usgs.gov .