Each of the available transforms creates an output dataset where each output band is a linear combination of all the input bands. Follow these steps to continue:

  1. From the Transform Method drop-down list, select a method. The choices are as follows:
    • Independent Component Analysis: The ICA method works well with hyperspectral data because it is more likely to treat sparse targets as important features, compared with the PCA or MNF methods. However, ICA can take a significantly longer time to process.
  2. Click Transform Params. The ICA Parameters dialog appears. See Independent Components Analysis for more information on the parameters required for ICA. Reducing the Number of Output IC Bands value will result in faster processing, but at the risk of omitting some subtle features. ICA is a memory-intensive process. If you receive a memory allocation error during processing, reduce the X Sampling and Y Sampling values to reduce the memory usage. However, this can potentially change the statistics such that you miss sparse targets.

    • Principal Component Analysis: The PCA method creates a number of PC bands, which are linear combinations of the original spectral bands that are uncorrelated. You can calculate the same number of output PC bands as input spectral bands. The first PC band contains the largest percentage of data variance and the second PC band contains the second largest data variance, and so on.

    Click Transform Params. The PCA Parameters dialog appears. See Principal Components Analysis for more information on the required parameters for PCA.

    • Minimum Noise Fraction: The MNF transform is a linear transformation that uses separate PCA rotations to segregate noise in the data and to reduce the dimensionality of the original dataset.
    • Click Transform Params. The MNF Parameters dialog appears. Refer to Minimum Noise Fraction Transform for more information about the required parameters for MNF.

  3. The use of masks is recommended along with image transforms in cases where the image contains large areas that you are not interested in processing (such as water or black-fill pixels). In the Parameters dialog for each image transform method is a Mask Information section. From the Masking Method drop-down list, select one of the following options:
    • None: Select this option if you do not want to use a mask.
    • Mask user-specified value: Select this option to mask out certain pixels within an image. In the Fill Value field, enter the pixel value for the pixels you want to exclude from processing. When you apply a mask, the mask is used to generate transform coefficients. Yet the coefficients are applied to every pixel in the input image, whether a mask was used or not. Set the Zero Out Masked Pixels toggle button to Yes to apply the mask to the transform image so that masked values are 0.
    • User-defined mask file: Select this option to apply a separate mask file to the hyperspectral image. The mask file should be a binary image consisting of values of 0 and 1, and it must have the same spatial extent and projection as the hyperspectral image. Click Choose, and select a mask file. Set the Zero Out Masked Pixels toggle button to Yes to apply the mask to the transform image so that masked values are 0.
  4. Click Run Transform. When processing is complete, the results appear in the Dimensionality Reduction and Band Selection panel.
  5. Review the transformed image bands to determine which bands contain useful information. While the example discussed here is for ICA results, the same applies to PCA and MNF.
  6. The ICA Results section of the Dimensionality Reduction and Band Selection panel lists every transformed band. Double-click a band or click Display Band to open the selected band in a display group. Click Animate Bands to launch ENVI's animation tool, which provides a convenient way to quickly review all the bands.

    For ICA, click Plot 2D Coherence to see a plot of the spatial coherence for each transformed band. For PCA and MNF, click Plot Data Variance and select a plot to view. The choices are: Plot Variance (ENVI), Plot Variance (Log axis, IDL), Plot Eigenvalues (ENVI), and Plot Eigenvalues (Log axis, IDL).

    Advanced users may explore transformed results further by clicking Plot Band Weights. The Band Weightings dialog appears. This dialog provides some insight into how much each input spectral band contributes to each transformed image band.

  7. Select the transformed bands you wish to keep by clicking one of the following buttons:
    • Select Graphically: The THOR Spectral Subsetting dialog appears.
    • Select by List: The File Spectral Subset dialog appears.

THOR will automatically apply the selected transform to the target and background (if any) signatures before spectral matching is performed later in your workflow.