Visual Classification Guide
The Visual Classification feature is used to recognize different textures, different orientations, and other features of the same Target Object. This document mainly describes how to adjust related vision parameters after enabling Visual Classification.
1. Operation Guide
When the same Target Object in the actual scene has different textures, different orientations, or other features, select the Visual Classification feature when creating a new Project; if it was not selected when creating the new Project, go to the Task Information page, click Edit in the lower-right corner, and then select Visual Classification
On the Target Object page - Visual Classification, click Enter Type to open the Type Recorder
Enter types
Click + to add a type
Enter the type id
Capture an image
Draw a box around the corresponding image region in the image. Enter at least 10 samples for each type
Adjust parameters
| Parameter | Description | Default Value | Range | Tuning Recommendation |
|---|---|---|---|---|
| Scale Ratio | Scales the images in the entered type database and extracts image features. The smaller the Scale Ratio, the faster the category information is computed. | 1.0 | 0.1 - 2.0 | 1.0 |
| Augment Data by Rotation | Rotates the entered type images to augment the number of images in the type database | Selected | / | / |
| Image Rotation Count | After selecting Augment Data by Rotation, the number of images in the database is increased according to the Image Rotation Count. For example, if the Image Rotation Count is 4, the image is rotated counterclockwise by 90° (360°/4) each time, for a total of 4 rotations; if the Image Rotation Count is 3, the image is rotated counterclockwise by 120° (360°/3) each time, for a total of 3 rotations | 2 | [1,360] | The larger the value, the more rotation angles are added to the augmented images |
The Scale Ratio affects the Takt Time of visual classification. The smaller the Scale Ratio, the faster the Takt Time. Under normal circumstances, the Takt Time of visual classification is within 0.5s
Finally, click Compute Classification Parameters and close the Type Recorder
Return category information
Click Robot Panel to open Robot Configuration. Under pick-related information, select Target Object Category so that the robot can obtain the relevant category information
2. Parameter Tuning Guide
After selecting each configuration item in Task Information, adjust parameters under Vision Parameters - 2D Recognition - Visual Classification
| Parameter | Description | Default Value | Range | Tuning Recommendation |
|---|---|---|---|---|
| Save Image Classifier Data | Saves the information entered in the Type Recorder when selected | Selected | / | Usually no adjustment is needed |
| Exclude Background Outside the Instance | When selected, the images detected by visual classification retain only the Mask region and remove the image Background, which can reduce the impact of the image Background on category judgment. | Not selected | / | Usually no adjustment is needed |
| Classification Mode |
| Target Object/Scene Feature Classification Mode |
| When the Target Object and Scene are similar and differ only in color, use Color Classification Mode. In all other cases, use the default Target Object/Scene Feature Classification Mode |
Example: When the Target Object/Scene features are obvious, use the Target Object/Scene Feature Classification Mode for classification.
Example: When color features are obvious, use the Color Classification Mode for classification.
