To customize the pre-trained TCA Component Cleanliness Object Classification Model, you extend the provided ML (machine learning) ML based model by retraining with your own acquired images saved in a conventional TCA job run. To do so, export the archived image to your local file system.

NOTICE

notice

Existing images in a pre-trained model

Do not remove existing images in a pre-trained ML (machine learning) based model, because the training information added with the corresponding image will be deleted from the model.

  1. The TCA Object Classification images are installed using the Microscopy Installer.
  2. The Manage Templates mode is selected.
  1. From the Show drop-down list, select Intellesis Object Classification Models.
  2. The models are displayed.
  3. Select the desired TCA Object Classification model, and click Copy and Edit .
  4. The Intellesis Object Classification workbench is displayed showing the mandatory particle type classes for TCA.
  5. Click Import from Archive or Import Images if you want to import the image from your local disc.
  6. The browser opens.
  7. Select the exported image from the file system for training, and click Open. Be aware that the images you use for training were acquired with the same channel combination that is used in the selected model.
  8. The image and a table containing the IDs of the objects in the image are displayed in the Center Screen Area.
  9. In the classes list, select a class, and in the image, click on an object that belongs to this particle type class. Repeat this for a certain number of particles.
  10. You have labeled the object and assigned it to the selected class indicated by the same color of the corresponding particle type class.

  11. Click Train & Classify.
  12. The model is retrained based on the labeling. A prediction is displayed in the table.
  13. Review the predictions. To do so, in the table, sort the Prediction column table class and then click a predicted object in the table. Check that the algorithm has correctly labeled the object in the image. If you are not sure whether an object is metallic-shiny or not, on the Display tab, activate Single Channel, and activate Pol-135. The metallic-shiny particles will appear with white areas. By selection of the Pol-90 channel, metallic particles must appear black throughout.
    If the results are not precise enough, label additional objects. Click again Train & Classify.
  1. The model is retrained based on the labeling. The model is available at Manage Templates > Intellesis Object Classification Models, e.g. Model Name (1). To change the model name, in the Properties area, click , and change the name.

Note that you must select the retrained model again in the TCA job template > Intellesis Object Classification workbench > Intellesis Object Classification tool.