Reduce Spectral Bands (Image Analyst Tools)

Summary

Reduces the number of spectral bands in multiband imagery using Minimum Noise Fraction (MNF) and Principal Components Analysis (PCA) analytical methods.

Usage

Both PCA and MNF are widely used to reduce the number of bands in hyperspectral imagery, but they differ in what they optimize, how they treat noise, and how suitable they are for hyperspectral data.
PCA is a statistical transform that reprojects the original spectral bands into a new set of orthogonal components, ordered by the total variance explained. The first component explains the largest variance (usually containing the most information), the second component explains the next largest amount of variance, and so on.
MNF is noise-adjusted PCA, specifically designed for hyperspectral imagery. It estimates the noise covariance matrix, and converts the data so noise has unit variance (noise is normalized so that one unit of variation equals one standard deviation of noise) and is uncorrelated. Once noise variance = 1, and noise correlation = 0, then any direction in spectral space with variance > 1 must contain signal. Then the data is transformed by PCA and components ordered by decreasing Signal-to-Noise Ratio (SNR), separating meaningful data from noise for better analysis, unlike standard PCA which prioritizes total variance.
The tool takes a multiband or hyperspectral input image and outputs a transformed raster with the specified number of MNF and PCA components (bands). Transforming and reducing spectral bands to those containing the most relevant image information increases processing efficiency and accuracy.

Label	Explanation	Data type
Input Raster	A multiband or hyperspectral image. If the input is a multidimensional raster, the first slice will be used.	Raster Dataset; Mosaic Dataset; Mosaic Layer; Raster Layer; Image Service
Output Raster	The transformed raster that contains the MNF or PCA components. The number of bands is equal to Number of Components parameter value.	Raster Dataset
Reduction Method (Optional)	Specifies the method used to perform the transformation. Minimum Noise Fraction (MNF)—The Minimum Noise Fraction (MNF) method will be used to reduce the number of bands. This is the default. Principal Components Analysis (PCA)—The Principal Components Analysis (PCA) method will be used to reduce the number of bands.	String
Number of Components (Optional)	The number of MNF or PCA components that the output will contain. The default value is equal to the number of bands contained in the image; however, if the image has more than 20 bands, then the default will be set to 20.	Long
Output MNF Transform (Optional)	The MNF transformation coefficients used to transform the input image in the form of number of input bands * Number of Components parameter value. You can use this raster dataset to transform another image using the same coefficients.	Raster Dataset
Output PCA Loadings (Optional)	A table listing weights that each input band contributes to the principal components. These weights define the correlations of the input data and the output principal components.	Table
Output PCA Eigenvalues (Optional)	The eigenvalues table which indicates the variance percentage of each component. Eigenvalues help you define the number of principal components that are needed to represent the dataset.	Table

ReduceSpectralBands(in_raster, out_raster, {method}, {num_components}, {out_mnf_transform}, {out_pca_loadings}, {out_pca_eigenvalues})

Name	Explanation	Data type
in_raster	A multiband or hyperspectral image. If the input is a multidimensional raster, the first slice will be used.	Raster Dataset; Mosaic Dataset; Mosaic Layer; Raster Layer; Image Service
out_raster	The transformed raster that contains the MNF or PCA components. The number of bands is equal to `num_components` parameter value.	Raster Dataset
method (Optional)	Specifies the method used to perform the transformation. `MNF`—The MNF method will be used to reduce the number of bands. This is the default. `PCA`—The PCA method will be used to reduce the number of bands.	String
num_components (Optional)	The number of MNF or PCA components that the output will contain. The default value is equal to the number of bands contained in the image; however, if the image has more than 20 bands, then the default will be set to 20.	Long
out_mnf_transform (Optional)	The MNF transformation coefficients used to transform the input image in the form of number of input bands * `num_components` parameter value. You can use this raster dataset to transform another image using the same coefficients.	Raster Dataset
out_pca_loadings (Optional)	A table listing weights that each input band contributes to the principal components. These weights define the correlations of the input data and the output principal components.	Table
out_pca_eigenvalues (Optional)	The eigenvalues table which indicates the variance percentage of each component. Eigenvalues help you define the number of principal components that are needed to represent the dataset.	Table

Code sample

ReduceSpectralBands example 1 (stand-alone script)

Reduce the number of bands using the MNF method.

# Import system modules   
import arcpy   

# Define variables 
input_raster = r"c:\data\hyperspectral_img.tif"
output_raster = r"c:\output\mnf_img.tif"
method = "MNF"
components = 10 
transform = r"c:\output\mnf_transform"

# Run ReduceSpectralBands
output = arcpy.ia.ReduceSpectralBands(in_raster=input_raster, out_components=output_raster, method=method, num_components=components, out_mnf_transform=transform)

ReduceSpectralBands example 2 (stand-alone script)

Reduce the number of bands using the PCA method.

# Import system modules   
import arcpy   

# Define variables 
input_raster = r"c:\data\hyperspectral_img.tif"
output_raster = r"c:\output\pca_img.tif"
method = "PCA"
components = 15
loadings = r"c:\output\loadings_table.csv"
eigenvalues = r"c:\output\eigenvalues_table.csv"

# Run ReduceSpectralBands
output = arcpy.ia.ReduceSpectralBands(in_raster=input_raster, out_components=output_raster, method=method, num_components=components, out_pca_loadings=loadings, out_pca_eigenvalues=eigenvalues)

Environments

Cell Alignment, Cell Size, Compression, Current Workspace, Extent, Geographic Transformations, NoData, Output CONFIG Keyword, Output Coordinate System, Parallel Processing Factor, Pyramid, Raster Statistics, Resampling Method, Scratch Workspace, Snap Raster, Tile Size

Licensing information

Basic: Requires Image Analyst
Standard: Requires Image Analyst
Advanced: Requires Image Analyst

References

"Green, A. A., Berman, M., Switzer, P., and Craig, M. D., 1988, A transformation for ordering multispectral data in terms of image quality with implications for noise removal: IEEE Transactions on Geoscience and Remote Sensing, v. 26, no. 1, p. 65-74."

Find a geoprocessing tool