spectralMatch

Identify unknown regions or materials using spectral library

Since R2020a

Syntax

score = spectralMatch(libData,hcube)

score = spectralMatch(libData,reflectance,wavelength)

score = spectralMatch(___,Name,Value)

Description

score = spectralMatch(libData,hcube) identifies regions in a hyperspectral data cube by matching spectra signature of each pixel to the spectral data read from the ECOSTRESS spectral library libData.

score = spectralMatch(libData,reflectance,wavelength) identifies a region or material by matching its spectral reflectance values, specified as reflectance and wavelength, with the values available in the ECOSTRESS spectral library libData.

score = spectralMatch(___,Name,Value) specifies options using one or more name-value pair arguments in addition to any combination of input arguments in previous syntaxes.

Note

This function requires the Hyperspectral Imaging Library for Image Processing Toolbox™. You can install the Hyperspectral Imaging Library for Image Processing Toolbox from Add-On Explorer. For more information about installing add-ons, see Get and Manage Add-Ons.

The Hyperspectral Imaging Library for Image Processing Toolbox requires desktop MATLAB^®, as MATLAB Online™ or MATLAB Mobile™ do not support the library.

Examples

Segment Vegetation Regions in Hyperspectral Data Using Spectral Matching

The spectral matching method compares the spectral signature of each pixel in the hyperspectral data cube with a reference spectral signature for vegetation from an ECOSTRESS spectrum file.

Read the spectral signature of vegetation from the ECOSTRESS spectral library.

filename = "vegetation.tree.tsuga.canadensis.vswir.tsca-1-47.ucsb.asd.spectrum.txt";
libData = readEcostressSig(filename);

Read the hyperspectral data into the workspace.

hcube = hypercube("paviaU.hdr");

Compute the distance scores of the spectrum of the hyperspectral data pixels with respect to the reference spectrum.

score = spectralMatch(libData,hcube);

Display the distance scores. The pixels with low distance scores are stronger matches to the reference spectrum and are more likely to belong to the vegetation region.

imagesc(score)
colorbar

Define a threshold for detecting distance scores that correspond to the vegetation region.

threshold = 0.3;

Create a binary image by applying the threshold. The regions in the binary image with a value of 1 correspond to the vegetation regions in the data cube with distance scores less than the threshold. All other pixels have a value 0.

bw = score < threshold;

Segment the vegetation regions of the hyperspectral data cube by using the indices of the maximum intensity regions in the binary image.

T = reshape(hcube.DataCube,[size(hcube.DataCube,1)*size(hcube.DataCube,2) size(hcube.DataCube,3)]);
Ts = zeros(size(T));
Ts(bw == 1,:) = T( bw==1 ,:);
Ts = reshape(Ts,[size(hcube.DataCube,1) size(hcube.DataCube,2) size(hcube.DataCube,3)]);

Create a new hypercube object that contains only the segmented vegetation regions.

segmentedDataCube = hypercube(Ts,hcube.Wavelength);

Estimate the RGB colour image of the original data cube and the segmented data cube by using the colorize function.

rgbImg = colorize(hcube,Method="rgb",ContrastStretching=true);
segmentedImg = colorize(segmentedDataCube,Method="rgb",ContrastStretching=true);

Overlay the binary image on the RGB version of the original data cube by using the imoverlay function.

B = imoverlay(rgbImg,bw,"Yellow");

Display the RGB colour images of the original data cube and the segmented data cube along with the overlaid image. The segmented image contains only the vegetation regions that are segmented from the original data cube.

figure
montage({rgbImg,segmentedImg,B},Size=[1 3])
title("Original Image | Segmented Image | Overlaid Image")

Identify Unknown Spectral Signature Using Spectral Matching

Read reference spectral signatures from the ECOSTRESS spectral library. The library consists of 15 spectral signatures belonging to manmade materials, soil, water, and vegetation. The output is a structure array that stores the spectral data read from ECOSTRESS library files.

dirname = fullfile(matlabroot,'toolbox','images','supportpackages','hyperspectral','hyperdata','ECOSTRESSSpectraFiles');
libData = readEcostressSig(dirname);

Load a .mat file that contains the reflectance and the wavelength values of an unknown material into the workspace. The reflectance and the wavelength values together comprise the test spectrum.

load spectralData 'reflectance' 'wavelength'

Compute the spectral match between the reference spectrum and test spectrum using spectral information divergence (SID) method. The function computes the distance score for only those reference spectra that have bandwidth overlap with the test spectrum. The function displays a warning message for all other spectra.

score = spectralMatch(libData,reflectance,wavelength,'Method','SID');

Display the distance scores of the test spectrum. The pixels with lower distance scores are stronger matches to the reference spectrum. A distance score value of NaN indicates that the corresponding reference spectrum and the test spectrum do not meet the overlap bandwidth threshold.

score

score = 1×15

  297.8016  122.5567  203.5864  103.3351  288.7747  275.5321  294.2341       NaN       NaN  290.4887       NaN  299.5762  171.6919   46.2072  176.6637

Find the minimum distance score and the corresponding index. The returned index value indicates the row of the structure array libData that contains the reference spectrum that most closely matches a test spectrum.

[value,ind] = min(score);

Find the matching reference spectrum by using the index of the minimum distance score, and display the details of the matching spectral data in the ECOSTRESS library. The result shows that the test spectrum match most closely with the spectral signature of sea water.

matchingSpectra = libData(ind)

matchingSpectra = struct with fields:
                     Name: "Sea Foam"
                     Type: "Water"
                    Class: "Sea Water"
                 SubClass: "none"
             ParticleSize: "Liquid"
                    Genus: [0×0 string]
                  Species: [0×0 string]
                 SampleNo: "seafoam"
                    Owner: "Dept. of Earth and Planetary Science, John Hopkins University"
          WavelengthRange: "TIR"
                   Origin: "JHU IR Spectroscopy Lab."
           CollectionDate: "N/A"
              Description: "Sea foam water. Original filename FOAM Original ASTER Spectral Library name was jhu.becknic.water.sea.none.liquid.seafoam.spectrum.txt"
              Measurement: "Directional (10 Degree) Hemispherical Reflectance"
              FirstColumn: "X"
             SecondColumn: "Y"
           WavelengthUnit: "micrometer"
                 DataUnit: "Reflectance (percent)"
              FirstXValue: "14.0112"
               LastXValue: "2.0795"
          NumberOfXValues: "2110"
    AdditionalInformation: "none"
               Wavelength: [2110×1 double]
              Reflectance: [2110×1 double]

Plot the reflectance values of the test spectrum and the corresponding reference spectrum. For the purpose of plotting and visualizing the shape of the reflectance curves, rescale the reflectance values to the range [0, 1] and interpolate test reflectance values to match the reference reflectance values in number.

figure
testReflectance = rescale(reflectance,0,1);
refReflectance = rescale(matchingSpectra.Reflectance,0,1);
testLength = length(testReflectance);
newLength = length(testReflectance)/length(refReflectance);
testReflectance = interp1(1:testLength,testReflectance,1:newLength:testLength);

plot(refReflectance)
hold on
plot(testReflectance,'r')
hold off
legend('Matching reference reflectance','Test reflectance')
xlabel('Number of samples')
ylabel('Reflectance value')

Input Arguments

libData — Spectral signature from library

structure | structure array

Spectral data from spectral library, specified as a structure or a 1-by-K structure array. K is the number of spectral signatures from the library to be used for spectral matching. Each spectral library structure must contain Reflectance and Wavelength fields consisting of numeric vectors of equal length, such as a spectral signature read from the ECOSTRESS library using the readEcostressSig function.

hcube — Input hyperspectral data

hypercube object

Input hyperspectral data, specified as a hypercube object. The DataCube property of the hypercube object contains the hyperspectral datacube.

reflectance — Reflectance values

C-element vector

Reflectance values, specified as a C-element vector. C is the number of wavelengths for which the reflectance values have been measured.

wavelength — Wavelength values

C-element vector

Wavelength values, specified as a C-element vector. C is the number of wavelengths for which the reflectance values have been measured.

Name-Value Arguments

Specify optional pairs of arguments as Name1=Value1,...,NameN=ValueN, where Name is the argument name and Value is the corresponding value. Name-value arguments must appear after other arguments, but the order of the pairs does not matter.

Before R2021a, use commas to separate each name and value, and enclose Name in quotes.

Example: spectralMatch(libData,hcube,'MinBandWidth',0.5)

Method — Spectral matching method

'sam' (default) | 'sid' | 'sidsam' | 'jmsam' | 'ns3'

Spectral matching method, specified as the comma-separated pair consisting of 'Method' and one of these values:

• 'sam' — Spectral angle mapper (SAM) method, which measures the similarity between two spectra by computing the angular distance between them.

• 'sid' — Spectral information divergence (SID) method, which measures the similarity between two spectra by computing the difference between their probability distribution values.

• 'sidsam' — Mixed spectral similarity method, which measures the similarity between two spectra by combining the SID and SAM distance measures.

• 'jmsam' — Jeffries Matusita-Spectral Angle Mapper (JMSAM), which measures the similarity between two spectra by combining the Jeffries Matusita (JM) and SAM distance measures.

• 'ns3' — Normalized spectral similarity score (NS3) method, which measures the similarity between two spectra by combining the Euclidean and SAM distance measures.

For details about these spectral matching methods, see Algorithm.

Data Types: char | string

MinBandWidth — Minimum overlap bandwidth

300 (default) | positive scalar

Minimum overlap bandwidth, specified as the comma-separated pair consisting of 'MinBandWidth' and a positive scalar in nanometers. The overlap bandwidth between the reference spectrum and the test spectra is defined as:

BW_overlap = W_max − W_min

W_min is the maximum of minimum wavelengths in the reference and test spectra.

W_max is the maximum of maximum wavelengths in the reference and test spectra.

The 'MinBandWidth' argument defines the minimum expected value for the overlap bandwidth between the spectral values of the test material and the ECOSTRESS spectral data.

Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64

Output Arguments

score — Distance scores

3-D numeric array | matrix | K-element column vector | scalar

Distance scores, returned as a 3-D numeric array, matrix, K-element column vector, or scalar. The dimensions of the output score depend on the dimensions of the libData and whether the test data is a hypercube object or a wavelength and reflectance pair.

If the test spectral signatures are specified as a hypercube object, hcube and the data cube is of size M-by-N-by-C:

Dimension of input argument, libData	Dimension of output, score
1-by-K, containing K reference signatures read from K number of spectrum files	3-D numeric array of size M-by-N-by-K containing the distance score for each pixel with respect to K reference signatures The values in each channel of K are the distance scores of the spectra of each pixel with respect to the spectral data in the corresponding row of libData. Similarly, the values in the second channel relate to the spectral data in the second row of libData.
1-by-1, containing reference signature read from one spectrum file (K = 1)	matrix of size M-by-N, The matrix contains the distance score for each pixel's spectra with respect to a reference signature.

If the test spectral signature is specified as reflectance and wavelength values:

Dimension of input argument, libData	Dimension of output, score
1-by-K, containing K reference signatures read from K number of spectrum files	K-element vector containing the distance score of the test spectra with respect to K reference signatures. Each element of the vector is the distance score of the test reflectance values with respect to the spectral data in the corresponding row of libData.
1-by-1, containing reference signature read from one spectrum file (K = 1)	scalar

Data Types: double

Limitations

This function does not support parfor loops when the Method is specified as "sam", "sid", "jmsam", or "ns3", as its performance is already optimized. (since R2023a)

Algorithms

Spectral Angle Mapper (SAM)

For the "sam" method, the spectralMatch function computes the SAM score α using this formula.

$$\alpha ={\cos }^{-1}\left(\frac{{\displaystyle \sum _{i=1}^C{t}_i{r}_i}}{\sqrt{{\displaystyle \sum _{i=1}^C{t}_i^2}}\sqrt{{\displaystyle \sum _{i=1}^C{r}_i^2}}}\right).$$

where, r and t are the reference and test spectra, respectively. r_i and t_i are the i^th elements of the vectors r and t, respectively. C is the length of vectors r and t.

Spectral Information Divergence (SID)

For the "sid" method, the spectralMatch function normalizes the reference spectra refSpectra and test spectra testSpectra and computes the SID value using this formula.

$$SID={\displaystyle \sum _{i=1}^C{p}_i\log \left(\frac{p_i}{q_i}\right)+}{\displaystyle \sum _{i=1}^C{q}_i\log \left(\frac{q_i}{p_i}\right)}.$$

q and p are the vectors of normalized reference and test spectra, respectively. q_i and p_i are the i^th elements of the vectors q and p, respectively. C is the length of vectors q and p.

SID-SAM

For the "sidsam" method, the spectralMatch function computes the SID-SAM value using this formula.

$$SIDSAM=SID\times \tan \left(\alpha \right)$$

SID is the SID value, and α is the SAM score.

Jeffries Matusita-Spectral Angle Mapper (JMSAM)

For the "jmsam" method, the spectralMatch function computes the JMSAM score using this formula.

$$JMSAM=J{M}_{distance}\times \tan \left(\alpha \right)$$

JM_distance is the JM distance, and α is the SAM score. This method computes the JM distance using this formula.

$$J{M}_{\text{distance}}=2\left(1-e^{-B}\right)$$

B is the Bhattacharyya distance:

$$B=\frac{1}{8}{\left({\mu }_t-{\mu }_r\right)}^T{\left[\frac{{\sigma }_t+{\sigma }_r}{2}\right]}^{-1}\left({\mu }_t-{\mu }_r\right)+\frac{1}{2}\ln \left[\frac{\left|\frac{{\sigma }_t+{\sigma }_r}{2}\right|}{\sqrt{\left|{\sigma }_t\right|\left|{\sigma }_r\right|}}\right]$$

μ_r and μ_t are the mean values of the reference and test spectra, respectively. σ_r and σ_t are the covariance values of the reference and test spectra, respectively.

Normalized Spectral Similarity Score (NS3)

For the "ns3" method, the spectralMatch function computes the NS3 score using this formula.

$$NS3=\sqrt{A_{Euclidean}{}^2+{\left(1-\cos \left(\alpha \right)\right)}^2}$$

A_Euclidean is the Euclidean distance, and α is the SAM score. This method computes the Euclidean distance using this formula.

$$A_{Euclidean}=\sqrt{\frac{1}{C}{\displaystyle \sum _{i=1}^C{\left(t_i-r_i\right)}^2}}$$

r and t are the vectors of reference and test spectra, respectively. r_i and t_i are the i^th elements of the vectors r and t, respectively. C is the length of vectors r and t.

Version History

Introduced in R2020a