| Description | Performs a support vector machine classification in the spectral dimension. |
| Input | Data Type = PixelBased/ObjectBased Value Type = FloatingPoint |
| Output | Data Type = (Input Data Type) Value Type = Classification
|
| Parameters | 
- Solver
- C-SVC: C-support vector classification.
- Classification Type
- One-Versus-One: Classifies groups pairwise.
- One-Versus-All: Classifies each group against the rest.
- C-Value: Violation regularization parameter. Large values represent a soft boundary, enhancing the significance of the violation magnitude in the optimization procedure, while small values represent a hard boundary where only the number of violation matters.
- Kernel Type: Kernel type for feature space transformation.
- Linear: (xT*x')
- Polynomial: (a*xT*x' + b)d
- Gaussian: exp(-a*|x-x'|2)
- Sigmoidal: tanh(a*xT*x'+b)
- Kernel Parameters
- a: Multiplication constant.
- b: Addition constant.
- d: Exponentiation constant.
- Classification Threshold: The maximum distance to the hypersurface for conducted classification to a specific group index. (The SVM classifies spectra depending on which side of the speparating hypersurfaces it (mostly) lies. Use this parameter to constrain this classification depending on the distance of the spectrum to the hypersurface.)
|
| Comments | - Supervised method which needs at least two group assignments.
- In multicategorial classification, a voting strategy is used (in the case of Classification Type = One-Versus-All, only the relevant group can get votes).
- The thresholding is applied after the voting, which means that a sample can only be classified to its member group resulting from the non-thresholded classification.
|
| Available | Version 2.4 |
