Machine Learning Classifier

Features:

Performs classification of 3D datasets using machine learning algorithms. Training is conducted using a single file.

Steps:

1.      Open an IDCube file.

2.      Click the Machine Learning tab and select Machine Learning Classifier from the menu.

The new window Machine Learning will open. The window will present the image in the same mode as the main interface (such as RGB as shown below). The window will also provide information about the dataset (left upper corner).

The toolbox enables the user to use two types of labeling files. Available options are:

• Single class labels. This option enables using multiple labels of the same class.

• Multiple class labels. This option enables using multiple classes with one label per class.

The algorithm also counts unlabeled part of the images as background class for both of these options, that the total number of classes for a single class label is equal to two: labeled + background.  

Note: IDCube treats the background data as a separate additional class. The labels of this class are identified automatically.

3.      Click Load Label Data and select a file previously saved from labeling. The file should have an .m (or .mat) extension and is automatically generated along with the png file by using the Create Label tool. The labeled file with two labeled areas corresponding to different leaves will appear in the right corner.

4.      Select a method from a dropdown menu.

5.      An example is given with the Naïve Bayes method.  After selecting this method, press Train a model button. The training prediction model will appear as an image. The classified leaves seem to reproduce the actual shapes. 

The produced Confusion Matrix can be opened in a separate window by clicking the green Zoom button.

6.      After the training is complete, press Classify new dataset button  and open, for example, the original (not cropped) Rose_Leaves file.

7.      The results from this example where three individual classes are identified: the labeled classes (class 1, light green, class 2, black), and the background (class 3). (Note: The background class is identified automatically).

8.      Press the Overlay button to put all classes together. Although having the same dimension of the training set and the new datasets are not critical, having the same dimension enables overlay function by pressing the corresponding button.