Classification Studio
Classification Studio provides supervised spectral classification of hyperspectral images using user-defined training regions.
IDCubePro 2026 - Classification Studio
Overview
Classification Studio provides supervised spectral classification of hyperspectral images using user-defined training regions.
The workflow is based on selecting one or more training ROIs, calculating representative reference spectra, classifying the hyperspectral image, and applying an optional similarity threshold.
The Original Image panel displays the current hyperspectral image used for selecting training regions.
The displayed image is obtained from the active IDCubePro image when available, with the hyperspectral datacube used as a fallback.
Click Add Class to draw a rectangular training ROI on the Original Image.
Each ROI defines one training class.
The mean spectrum of all valid pixels inside the ROI is calculated and stored as the reference spectrum for that class.
You may add multiple classes before running classification.
Each training ROI is displayed on the Original Image using a class-specific color and label.
Pixels containing incomplete or non-finite spectra are excluded from the class reference spectrum.
The Algorithm Selection panel determines the spectral classifier used by Classification Studio.
SAM compares each image spectrum with the class reference spectrum using the spectral angle between the two vectors.
In the current Classification Studio implementation, SAM is converted to a normalized angular similarity score:
1 = strongest spectral match 0 = weak or orthogonal spectral match Higher values therefore indicate better matches.
Spectral Match provides an alternative spectral similarity method when enabled by the current implementation.
Reserved for linear spectral unmixing classification.
Reserved for matched-filter classification.
Reserved for Constrained Energy Minimization classification.
After adding at least one training ROI, click Classify.
Classification Studio calculates a similarity score for every image pixel relative to the selected reference spectrum or class.
An initial preview threshold is automatically calculated from the classification score distribution.
The Classified Image panel displays the pixels that satisfy the current classification threshold.
The displayed result updates when classification is performed or when a new threshold is applied.
The histogram displays the distribution of classification similarity scores.
This can be used to inspect the separation between strong and weak spectral matches and to select a meaningful threshold.
The threshold controls which pixels are accepted as classified pixels.
For the normalized SAM similarity score, higher scores represent better matches.
is classified as a positive match.
Enter a threshold value and click Apply to regenerate the binary classification mask.
For normalized SAM scores, valid threshold values normally fall between 0 and 1.
For example, a threshold near 0.90 retains only relatively strong matches.
After thresholding, Classification Studio calculates spatial metrics for each connected classified region.
Each row of the metrics table represents one connected classified region.
Exports the current region metrics to an Excel workbook.
The exported table contains one row for each connected classified region.
Reset removes all training ROIs, class reference spectra, classified results, histogram data, threshold state, and metrics.
The original hyperspectral image remains loaded and displayed.
1. Open Classification Studio with a hyperspectral datacube loaded.
3. Draw a representative training ROI.
4. Add additional training classes if needed.
5. Select the classification algorithm.
7. Inspect the Classified Image and score histogram.
8. Adjust the threshold if needed.
10. Review the region metrics.
11. Export metrics to Excel if required.
12. Click Reset to start a new classification analysis.