The Transform Band Weights tool allows advanced users to visualize the contribution of the various spectral bands to each output band of an image transform. Image transforms such as Principal Component Analysis (PCA), Minimum Noise Fraction (MNF), and Independent Component Analysis (ICA), are commonly used to:

  • Reduce the dimensionality of the input dataset to reduce processing time
  • Focus processing on the information of interest within the input file
  • Remove noise

Each output band of a transformed image is a linear combination of every input image. The Transform Band Weights tool provides a convenient interface to see which input bands contributed to each output band, thus helping to identify those spectral bands that are most important for finding targets of interest, or which bands contribute the most noise.

Follow these steps to run the Transform Band Weights tool.

  1. Choose one of the following options to start the Transform Band Weights tool:
    • Click Plot Band Weights in the Dimensionality Reduction and Band Selection panel. The transform data file will automatically pass into the tool.
    • Select THOR > THOR Transform Band Weights from the Toolbox. The Select Statistics File dialog appears. Select a statistics file from the PCA, MNF, or ICA results. You will also be asked whether you would like to select the input data file that the transform was derived from so that its wavelength information can be used in plotting. Otherwise, band numbers will be used. If selecting the input file, its number of bands must match the number of bands used to form the transformed image bands stored in the statistics file.
  2. The Transform Band Weightings list appears. This list contains each transformed band from the statistics file. Click on one of the transformed bands in the list to display a plot window. The x-axis shows the input spectral band number (or wavelength). The y-axis shows the percentage that each input spectral band contributed to the selected transformed image band. The sum of all values in the plot is 100%. Larger values on the plot indicate that those bands contributed the most to the selected transformed band.

You can also select several transformed bands by clicking and dragging in the list, or by holding down the Ctrl or Shift key while selecting from the list. The resulting plot will be the mean of all the plots for the selected bands:

Transform band weights can reveal interesting information about the input dataset, as the following figures show:

Setting the first toggle button in the Transform Band Weightings dialog to Create new window will display a new plot window each time you select a new transform band. Otherwise, the currently active plot window will be used. Leaving the bottom toggle button at Replace existing plot will replace the previous band selection with the new selection. If you select Accumulate plots, the new selection will overlay with the previous selections in the same plot window: