- To explore the different properties of a hyperspectral imagery.
- To become familiarized with various tools that can be used to process hyperspectral imagery, especially in doing image analysis such as computing NDVI.
For this exercise, the sample data used was downloaded from APEX (Airborne Prism Experiment) Open Dataset. This covers the area of Baden, Switzerland taken during the first flight campaign in June 2011.
The properties of the sample hyperspectral imagery were examined using Envi. Envi 5.3 was used mainly because of its user-friendly interface compared to the Classic version.
Upon inspection of the HDR file, it was found out that the image contains 285 bands with spectral range from 0.4131 – 2.421 micrometers.
Below is the initial band configuration used to display the image in a natural color.
Several band configurations were also tried out to visualize the image differently.
For the succeeding part of the assignment, the band combination R: Band 88, G: Band 61, B: Band 4 was used to visualize and identify some of the features from the image. A spectral library was created containing the different features identified from the image such as water, building, grass, forest, and road.
Looking at the spectral profile, the signature of each feature can be visualized in terms of reflectance for different wavelengths. Water and road features have a relatively lower reflectance rate compared to the others since they absorbed most of the energy and less is being recorded by the sensor. In contrast, vegetation features such as forest and grassland have the highest reflectance rate especially in the NIR region of the electromagnetic spectrum.
NDVI is the most common measure used in remote sensing to identify the areas that has vegetation cover. In its simplest form, the calculation of NDVI can be done using the formula:
NDVI = (rNIR - rRed) / (rNIR + rRed)
In Envi, the Band Math function was used to calculate the NDVI by creating the formula and specifying the bands in the NIR region.
Band 82 (0.8009) and Band 49 (0.6688) were used in the equation and the result was viewed using a Density Slice.
In this view, the vegetation areas can be seen very clearly with orange from red colors. Creating the density slice evidently helps in visualizing the vegetation areas as compared to the single band rendering of the same image. Interestingly, even in the areas with buildings, a lot of vegetation areas (i.e. parks, grass) are still present. The water features (blue) can be easily identified as well, since it has a low NDVI value as a result of low reflectivity within the range of the two bands used. In contrast, buildings and other concrete surfaces can be seen here with the green color.
Looking at the values of the classes from the density slice, a threshold from 0.633275 (orange) to 0.977273 (red) can be considered vegetation areas.
In this assignment, we explored some of the properties of a hyperspectral imagery using Envi, as well as the tools that may help us in doing preliminary image inspection. We conducted a simple NDVI calculation as well using the Band Math function in the software. Given the multitude of bands available in this image as compared to a multispectral imagery, it can be used for several applications as well such as identifying presence of minerals, as well as determining vegetation health. However, this also sometimes poses challenge during the pre-processing since not all bands are needed for a particular application and additional band reduction procedures must be carried out first before the actual analysis.
