Lab 9: Hyperspectral Remote Sensing

May 11, 2018

Goal
Te main goal of this lab is to understand processing and identifying features on hyperspectral images and selected spectral processing basics. Imaging spectrometry , hyperspectral images, and selected spectral processing basics will be performed. FLAASH will be introduced, and used to atmospherically correct a hyperspectral image. Then an introduction in determining the overall state of different vegetation types from their reflectance properties will be done.
Methods
Basic analysis of spectral processing
The first part of this lab introduces basic analysis performed on hyperspectral images. First the image cup95_rd.int is utilized which contains 50 bands(1.99-2.48um) of JPL-calibrated AVIRIS radiance for the Cuprite Mining District in Nevada. First a color image is displayed utilizing bands 183,193,and 207 in that order into the RGB combination. The spectral means of minerals within the area were plotted (Figure 1). Then mean spectra from ROIs was compared to spectral libraries (Figure 2).
Animation
Next an animation of the data was produced. Annimating the greyscale image allows for change to be seen more distinctly. spectral differences are more easily observed.

Atmospheric correction using FLAASH
Then FLAASH was used with an AVIRIS image JPL processed, and it contains calibrated at-sensor radiance values that were scaled to 2-byte signed integers. Then this image was atmospherically corrected. this process could not be performed without a license to FLAASH.

Luckily a corrected image was provided, when Bringing this into ENVI, the reflectance image, water vapor image, and cloud classification map are all available to be viewed. Then Spectral profiles of the area were observed (Figure 3-5). FLAASH flags bad bands that are not useful, and they can be seen in the three figures where the histogram line is broken. those broken areas contain the bad bands. FLAASH flagged 204 of 224 as bad bands. The three figures 3 4 and 5 show Urban, Vegetation, and Water land covers, and are compared to observe how atmospheric correction was performed.
Vegetation analysis
Then a vegetation analysis is performed utilizing the JaperRidge98av_flaash_refl.img bands 53,29 and 19 are loaded respectively into their RGB spots, The scene will be analyzed for forest fire risk using The agricultural stress tool (Figure 6), Fire Fuel tool (Figure 7), and forest health tool (Figure 8) are all used to show areas with high risk of fire.

Hyperspectal transformations
A minimum noise fraction transformation is used to determine dimensionality of image data, it divides noise in the data, and limits computational requirements for subsequent processing.EFFORT is a correction method which removes "saw-tooth" or calibration noise. Then MNF is used as a linear transformation, first noise covariance matrix is used to decorrelate and rescale noise in the data. This gives the noise a unit variance, and no band to band correlations this is called noise whitening. The second part uses the components of the original data after noise whitened, the dimensionality of the data is determined by looking at the final eigenvalues. In short noise is separated from the data improving spectral processing. MNF is used to remove noise. Figure 9 shows the MNF eigenvalues chart, where band 4 is where the plot starts flattening, and this is where the salt and pepper effect begins..

Results

We first observed minerals contained in the earths surface (Figure 1) and then compared them to spectral library in Figure 2. The Means in the image do not match up directly to the means taken from the spectral library.

Figure 1: Mean Spectra of Calcite, Budding, Varnish, Alunite, Kaolinite, silica, and Playa.

Figure 2: Comparing Spectral ROIs to spectral library.

Then spectral profile of urban (Figure 3), vegetation (Figure 4), and water (Figure 5) are observed with bad bands removed. The bands that are removed are bands that didnt have enough variability to differentiate different features.

Figure 3: Urban spectral profile with bad bands removed on the right.

Figure 4: Vegetation spectral profile with bad bands removed on the right.

Figure 5: Water spectral profile with bad bands removed on the right.

Vegetation analysis using agricultural stress tool (Figure 6), fire fuel (Figure 7), and Forest health tool (Figure 8), are viewed and can be utilized in assessing forest fire danger. Agricultural stress shows areas of crop stress. This tool uses Greenness, canopy water content, canopy nitrogen, light use efficiency, and leaf pigments to determine agricultural stress. Fire fuel tool creates a spatial map showing the distribution of fire fuels which are burn hazards, darker red colors are areas of high risk. this tool uses greenness, canopy water content and dry carbon to determine the risk. Finally the forest health tool spatialy maps overall health of forests. low stress forests generally have healthier vegetation, and high stress forests have less healthy vegetation. the forest health looks at greenness of vegetation, leaf pigments, water content, and light use efficiency to take into consideration the health of a forest.