Selected Hyperspectral Mapping Method Mirza Muhammad Waqar Contact: mirza.waqar@ist.edu.pk +92-21-34650765-79 EXT:2257 RG712 Course: Special Topics in Remote Sensing & GIS

Outlines Hyperspectral Data Hyperspectral vs Multispectral Data Analysis Hyperspectral Mapping Techniques Spectral Angle Mapper Matched Matching Spectral Feature Fitting Binary Encoding (BE) Complete Linear Spectral Unmixing Match Filtering

Revision – Hyperspectral Thematic Mapping Imaging Spectrometry Multispectral versus Hyperspectral Hyperspectral Image Acquisition Extraction of information from Hyperspectral data Preprocessing of Data Subset Study Area Initial Image Quality Assessment Visual Individual Band Examination Visual Examination of Color Composite Animation Statistical Individual Band Examination Radiometric Calibration In situ data Radiosounder Radiative Transfer based Atmospheric Correction DN Value Radiance Irradiance Apparent Reflectance (Albedo) Reflectance

Revision – Hyperspectral Thematic Mapping Selected Atmospheric Correction Models Flat Field Correction Internal Average Relative Reflectance (IARR) Empirical Line Calibration Reducing Data Redundancy Principal Component Transformation Minimum Noise Fraction Transformation (MNF) Endmember Determination Pixel Purity Index (PPI) n-dimensional visualization of endmembers in feature space Hyperspectral Mapping Method Spectral Angle Mapper (SAM)

Hyperspectral Data In order to be considered a specific data as hyperspectral, three conditions should be satisfied. Multiple bands High spectral resolution (i.e. narrowness of each band) Contiguity of bands. Landsat ASTER MODIS AVIRIS Hyperion

Signal-to-noise ratio Atmospheric interference Hyperspectral vs. Multispectral Data Analysis Hyperspectral Multispectral Bands Contiguous each other Discrete each other Analysis objectives Discriminate material among various earth surface features Categorize features Signal-to-noise ratio Lower (i.e. tendency of more noise) Higher Atmospheric interference More susceptible Less susceptible Analysis methods More reliance on physical and biophysical models More reliance on statistical techniques (ex. maximum likelihood classification) Hyperspectral data analysis is different from multispectral data analysis in some aspects. First, bands are contiguous in hyperspectral data, but not necessarily contiguous in multispectral data. Second, hyperspectral data analysis allows us to identify the unique materials present in the scene and map them throughout the image. With multispectral data, analysis is mostly limited to the categorization of pixels based on earth features. Third, more sensors bring difficulty in sensor calibration and behavior; therefore, hyperspectral data tend to contain more noise than multispectral data and they are more susceptible to atmospheric interference. Finally , hyperspectral data analysis relies more on physical and biophysical models, while multispectral data analysis relies on statistical techniques such as the maximum-likelihood classification method.

Multispectral vs Hyperspectral Mapping Multispectral Analysis methods are generally inadequate when applied to hyperspectral data: Inefficient: Multispectral methods are too computationally intensive when applied to high dimensional data Accuracy degradation Classification accuracy can actually decrease with the addition of extra bands that do not contribute meaningful information content. Loss of subtle detail The standard multispectral pattern recognition methods ultimately equate variance with information, which often results in subtle spectral variations being lost in the noise. So in turn for Hyperspectral mapping we need such mapping methods that focus on spectral contents and such decisions rules that identify the spectral characters that exists in the data.

Hyperspectral Mapping Techniques Atmospheric Correction Classification and target identification Whole pixel method Spectral Angle Mapper Spectral Feature Fitting Subpixel method Complete Linear Spectral Unmixing Matched Filtering Others Neural network Decision boundary feature extraction (DBFE)

Spectral Angle Mapper (SAM)

Spectral Angle Mapper (SAM) SAM compares test image spectra to a known reference spectra using the spectral angle between them. This method is not sensitive to illumination since the SAM algorithm uses only the vector direction and not the vector length. Spectral angle mapper (SAM) is frequently used for identifying materials in hyperspectral images. SAM compares test image spectra to a known reference spectra using the spectral angle between them. This method is not sensitive to illumination since the SAM algorithm uses only the vector direction and not the vector length. The result of the SAM classification is an image showing the best match at each pixel. This method is typically used as a first cut for determining the mineralogy and works well in areas of homogeneous regions. As described in an earlier slide, The USGS and some more organizations maintain large spectral libraries, mostly composed of mineral and soil types, where image spectra can be directly compared. a = spectral angle between two spectra n = number of bands Ti = reflectance value of band i in the test spectra Ri = reflectance value of band i in the reference spectra

Continuum Removal A continuum is a mathematical function used to isolate a particular absorption feature for analysis (Clark and Roush, 1984; Kruse et al, 1985; Green and Craig, 1985). LC= Continuum Removed Spectra using library spectra L = Library Spectra C λ = Least Square fit factor

Matched Matching Spectral Feature Fitting (SFF): A least-squares technique. SFF is an absorption-feature-based methodology. The reference spectra are scaled to match the image spectra after continuum removal from both data sets. (e.g. Tetracorder) Examines absorption features Depth Shape Ex. Tetracorder by USGS http://speclab.cr.usgs.gov/tetracorder.html Another approach to matching target and pixel spectra is by examining specific absorption features in the spectra. An advanced example of this method, called Tetracorder, has been developed by the U.S. Geological Survey. In Spectral Feature Fitting the user specifies a range of wavelengths within which a unique absorption feature exists for the chosen target. The pixel spectra are then compared to the target spectrum using two measurements: 1) the depth of the feature in the pixel is compared to the depth of the feature in the target, and 2) the shape of the feature in the pixel is compared to the shape of the feature in the target (using a least-squares technique).

Spectral Feature Fitting (SFF) Where Rb is reflectance in band center Rc is reflectance in continuum at band center Use specific bands to search for individual features and estimate a relative concentration based on band depth. First generate a continuum-removed spectrum for a specific feature in order to compare it with library spectra and image-derived spectra. Convolve library spectra with spectral response of sensor to generate an estimate of image derived reflectance spectra (i.e., assumes some form of atmospheric inversion has been applied to image data).

Matched Matching Binary Encoding (BE): The binary encoding classification technique encodes the data and end member spectra into 0s and 1s based on whether a band falls below or above the spectrum mean. An exclusive OR function is used to compare each encoded reference spectrum with the encoded data spectra and a classification image produced.

Binary Encoding (BE) Compute spectral mean of a sample (pixel) Assign a 1 to bands equal or greater than mean and 0 to those less than mean. Do the same for reference (e.g. spectral library) spectra. Compare the pattern as a measure of similarity. Compute spectral mean Rm of sample (pixel) over a local waveband of interest Assign a 1 to bands equal or greater than mean and 0 to those less than mean: If R(λ) ≥ Rm assign a “1” If R(λ) < Rm assign a “0”

Binary Encoding (BE)

Linear vs Non-Linear Mixing

Complete Linear Spectral Unmixing Calculate the fractions of endmembers in each pixel Endmembers Spectrally unique surface materials Similar to fuzzy classification with multispectral data analysis Results An abundance image, and Membership images The sets of spectrally unique surface materials existing within a scene are often referred to as the spectral endmembers for that scene. Linear Spectral Unmixing exploits the theory that the reflectance spectrum of any pixel is the result of linear combinations of the spectra of all endmembers inside that pixel. A linear combination in this context can be thought of as a weighted average, where each endmember weight is directly proportional to the area of the pixel containing that endmember. If the spectra of all endmembers in the scene are known, then their abundances within each pixel can be calculated from each pixel’s spectrum. Unmixing simply solves a set of n linear equations for each pixel, where n is the number of bands in the image. The unknown variables in these equations are the fractions of each endmember in the pixel. To be able to solve the linear equations for the unknown pixel fractions it is necessary to have more equations than unknowns, which means that we need more bands than endmember materials. With hyperspectral data this is almost always true. The results of Linear Spectral Unmixing include one abundance image for each endmember. The pixel values in these images indicate the percentage of the pixel made up of that endmember. For example, if a pixel in an abundance image for the endmember quartz has a value of 0.90, then 90% of the area of the pixel contains quartz. An error image is also usually calculated to help evaluate the success of the unmixing analysis.

Complete Linear Spectral Unmixing

Matched Filtering Partial unmixing technique Originally developed to compute abundances of targets that are relatively rare in the scene. Matched Filtering “filters” the input image for good matches to the chosen target spectrum by maximizing the response of the target spectrum within the data and suppressing the response of everything else. One potential problem with Matched Filtering is that it is possible to end up with false positive results. Matched Filtering is a type of unmixing in which only user-chosen targets are mapped. Unlike Complete Unmixing, we don’t need to find the spectra of all endmembers in the scene to get an accurate analysis (hence, this type of analysis is often called a ‘partial unmixing’ because the unmixing equations are only partially solved). Matched Filtering was originally developed to compute abundances of targets that are relatively rare in the scene. If the target is not rare, special care must be taken when applying and interpreting Matched Filtering results. Matched Filtering “filters” the input image for good matches to the chosen target spectrum by maximizing the response of the target spectrum within the data and suppressing the response of everything else (which is treated as a composite unknown background to the target). Like Complete Unmixing, a pixel value in the output image is proportional to the fraction of the pixel that contains the target material. Any pixel with a value of 0 or less would be interpreted as background (i.e., none of the target is present). One potential problem with Matched Filtering is that it is possible to end up with false positive results.

Hyperspectral Data Acquisition Raw Radiance Data Spectral Calibration At-Sensor Spectrally Calibrated Radiance Spatial Pre-Processing and Geocoding Radiometrically and Spatially processed radiance image Atmospheric Correction, solar irradiance correction Geocoding reflectance image Feature Mapping Data analysis for feature mapping Absorption band characterization Spectral feature fitting Spectral Angle Mapping Spectral Unmixing Minral Maps

Questions & Discussion