Classification of categories

Classifies object into categories using one of the following models:

• PLS-DA

• SIMCA

• Machine Learning

• Curve Separation

For more information on how to train a model see any the classification of nuts tutorials: Nut classification

Parameters

Model

Select the model created for the classification of the samples.

Category

Which category to apply the segmentation.

Classification type

• Object average spectrum

  • Classifies the object by taking the average spectrum from all pixels included in the object.

• Pixel class majority

  • Classifies the object by the majority of the pixels in the object.

Weights

The weights are used to control the importance of the pixels in an object for the classification using pixel class majority. This can be used to, for example, down-weigh objects' edge pixels because they may not be as representative as the core of the object.

The weights are specified as a string of values separated by semi-colons: W1;W2...;Wn where

• W1 is the weight for pixels at the edge of the object.

• W2 can be any number of semi-colon separated values, each giving the weight of the layer of pixels one step inward from the previous layer, where each layer is one pixel wide.

• Wn is the rightmost value and describes the weight for all remaining pixels in the center of the object.

The values of the weights are relative to each other, so it doesn’t matter what values you use.

Some examples:

• The default value for weights, 2;5;10, implies the center of the object is most representative of the spectrum (because of the 10) the edge values with 5 and 2 at the very edge are less representative.

• The weight string 1;1;1 would make all pixels equally important.

• The weight string 20;10;5;1, specifies that the pixels at the edge of the object are most important, the pixels one pixel in from the edge are slightly less important, the following step even less so and the center of the object is the least important.

• The weight string 1;10, implies that the very outer edge of the object is not very representative of the spectrum, while all of the remaining pixels are more representative.

• The weights 0;1, imply that the edge pixels are entirely unrepresentative and should not be considered for classification.

Pixel prediction

✅ Includes prediction of each pixel and visualization of the pixel prediction on the object.

⬜ Do not include the prediction or visualization

Connection

Set parent class for hierarchical model.

Show Train/Test column in table

Adds an additional column in table showing if the object was part of Train or Test set

✅ Add Train/Test column

⬜ Do not add Train/Test column