Hello Planetary Pals,
This week’s blog post is a special episode, crafted not for the normal bi-weekly blogs but instead as a part of the Planetary Science Seminar course. Our assignment is to talk about one of the resources or tools that we use in our research. My research is based on remote sensing of Mars, so in my research I use a variety of GIS software to process and view my images. Recently I have been working with CaSSIS color images obtained by the European Space Agency’s ExoMars orbiter. Using these images, I aim to do spectral analysis of the Martian surface in the search of smectite clays in the region southeast of Valles Marineris. These images are in a slightly processed state when I obtain them from the ESA and I must polish them further before they become useable. To do this I use ENVI, the Environment for Visualizing Images developed by Harris Geospatial Solutions.
ENVI is an image processing and analysis software that can be used for many types of data including SAR, LiDAR, and spectral images. It can work with satellite and airborne images and accepts many different data formats. ENVI is especially useful when analyzing and processing spectral information, and is thus the perfect tool for my research.
When I first access a CaSSIS image, it cannot be loaded into ENVI right away. This is because the ISIS3 format that the images are in initially are incompatible with ENVI; ENVI has trouble reading the label file and cannot recognize the NIR-RED-PAN-BLU bands. To fix this, a python script can be run that creates a compatible header file and discerns the proper band bins that can be recognized and loaded into ENVI as a cube file.
Next, the cube file can be loaded into ENVI as an RGB composite of the NIR-PAN-BLU channels.
Next, we need to tell ENVI to identify the region of the image that has full color coverage. As seen in the bottom left of the above image, there are some areas lacking overlap of the filters; these areas appear as the blue, magenta, yellow, and green on either end. To only analyze the area with total overlap, and to crop out those areas that lack it, we apply a mask to the image that ignores any pixel without full overlap. Any area that is not in this area is then ignored. We can then assign this masked area as the “Region of Interest” that future processing will work on.
Next, a gain must be applied. Applying a gain amplifies the signal from a given filter by a certain multiplier.
Following this, a process known as dark subtraction must be done. Spectra can be affected by the atmospheric dust in the Martian atmosphere, which obscures the true ground spectra. This is not a true atmospheric correction, since we do not have a reliable model to base Martian atmospheric conditions on. The dark subtraction attempts to remove dark spots and shadows by identifying the minimum pixel value that represents a background signature from each band. By removing this background, we can get truer spectra of the surface.
Next, a suite of band ratios can be created by taking ratios between each of the filters. Band ratios help exaggerate a spectral signature, which can be correlated to true, ideal lab spectra.
Next, we must fix the BLU filter; the BLU filter is the noisiest filter due to there being very little in the way of BLU wavelengths on the Martian surface. The goal is to make a synthetic blue band. We work with the BLU/PAN ratio and will later remove the PAN component. A Lee filter is applied to the BLU/PAN band ratio. The Lee filter replaces a pixel by taking the standard deviation of the surrounding pixels. The Low Pass filter smooths the image by replacing a pixel with a new value of the average pixel values surrounding it. Then we use a band math technique to re-isolate the new filtered BLU image by taking (BLU/PAN)*(PAN). Lastly, the synthetic blue filter can be created by taking (FilteredBLU*2)-(0.3*PAN). To be honest, I’m not sure why this math works, but whatever it does, it creates a final, synthetic blue channel to use that has the best signal-noise ratio.
Next, I can stack all the ratios and the NIR-RED-PAN-SyntheticBlue filters into one by using the layer stacking tool. Then we reapply the initial mask over the region of interest and the final image is ready to analyze.
The spectral signature of the surface is now ready to be investigated. The spectral profile can be compared to carefully obtained laboratory spectra of minerals. If the spectral signature of the area in ENVI matches that of the lab spectra, we can have confidence that such a location contains that mineral, although using a proper, hyperspectral imager such as CRISM is helpful as an additional check.
Overall, ENVI is a rather straightforward, user-friendly software to use when processing and analyzing spectral data and is the prefect tool for my scientific needs at the moment.