From GIS to Remote Sensing
This post is about a major update for the Semi-Automatic Classification Plugin for QGIS, version 4.4.0.
-added function for Landsat pan-sharpening
-added tooltips for classes and macroclasses in ROI and Signature tables
-bug fixing
-added tooltips for classes and macroclasses in ROI and Signature tables
-bug fixing
The main new function is the pan-sharpening of Landsat 7 and 8 images in the SCP tab Pre processing.
Pan-sharpening is the combination of the spectral information of multispectral bands (MS), with 30m spatial resolution, with the spatial resolution of a panchromatic band (PAN), which is 15m, available in Landsat 7 and 8. The result is a multispectral image with 15m spatial resolution.
For more information about pan-sharpening please read this open access paper (Johnson, Tateishi, & Hoan, 2012) .
For more information about pan-sharpening please read this open access paper (Johnson, Tateishi, & Hoan, 2012) .
Pan-sharpened images are useful for photo-interpretation of Landsat images, and can be used as input for land cover classifications.
Considering the forthcoming availability of Sentinel-2 images (with 10m spatial resolution), pan-sharpening of Landsat images could be very useful for integrating data or performing land cover change studies, exploiting the vast Landsat archive, at the spatial resolution of 15m.
It is worth pointing out that the pan-sharpening process alters the spectral signature of pixels, therefore classifications can include several errors due to spectral variability.
Considering the forthcoming availability of Sentinel-2 images (with 10m spatial resolution), pan-sharpening of Landsat images could be very useful for integrating data or performing land cover change studies, exploiting the vast Landsat archive, at the spatial resolution of 15m.
It is worth pointing out that the pan-sharpening process alters the spectral signature of pixels, therefore classifications can include several errors due to spectral variability.
I have implemented a Brovey Transform, where the pan-sharpened values of each multispectral band are calculated as (Johnson, Tateishi, Hoan, 2012) :
MSpan = MS * PAN / I
Where I is Intensity, which is a function of multispectral bands.
I have defined the following weights for I basing on several tests performed with the SCP. For Landsat 8, Intensity is calculated as:
I = (0.42 * Blue band + 0.98 * Green band + 0.6 * Red band ) / 2
For Landsat 7, Intensity is calculated as:
I = (0.42 * Blue band + 0.98 * Green band + 0.6 * Red band + NIR band) / 3
Download a sample image from here (data available from the U.S. Geological Survey), which is a subset of a Landsat 8 image. You can test the new function in the tab Landsat of the Pre processing:
- select the directory containing Landsat bands (and the metadata file);
- check the checkbox Perform pan-sharpening (Landsat 7 or 8);
- click Perform conversion and select the output directory.
The result of the pan-sharpening is a multispectral image with the spatial resolution of 15m. The example in the following figures shows the improvement of the spatial resolution.
Pan-sharpening seems to work well also with the DOS1 correction.
 |
| Original Landsat (30m) |
 |
| Pan-sharpened Landsat (15m) |
 |
| Original Landsat color composite RGB=543 (30m) |
 |
| Pan-sharpened Landsat color composite RGB=432 (15m) |
Please, remember that a Facebook group and a Google+ Community are available for sharing information and asking for help about the Semi-Automatic Classification Plugin.
References:
Johnson, B. A.; Tateishi, R. & Hoan, N. T. (2012). Satellite Image Pansharpening Using a Hybrid Approach for Object-Based Image Analysis ISPRS International Journal of Geo-Information, 1, 228 http://www.mdpi.com/2220-9964/1/3/228