Course Project: AT319 Lab 11 Classification Analysis

I. Intro

Date: 04/05/2023

During the first lab, we trained AI to identify the objects on the ground and let it calculate the area displaced on the surface. 

Part 1: Extract Spectral Bands

II. Extract Spectral Bands

1. Create a new project and load the database into ArcGIS Pro.
2. Drag the layer into contents pane. 
3. Select the layer, click Imagery in the tab -> Raster Function (in the Analysis section)
4. Search for “Extract Bands” and open it.
5. Set the “Raster” to the layer you chose, set the method to Band IDs. 
6. Set the combination to “4 1 3”, so red color is showing NIR, green color is showing red, blue color is showing blue.
7. Set the missing Band Action to Best Match. 
8. Rename the new layer, and create it. 

Part 2: Classification

III. Configure

Select the extracted band, click imagery in the tab, and click Classification Wizard

1. Set the “Classification Method” to Supervised.
2. Set the “Classification Type” to Object based
3. Set the “Classification Schema” to Use default schema
4. Rename the output
5. Next

IV. Segmentation

1. Set the “Spectral detail” to 8. Higher value will let more similar pixels to be grouped together.
2. Set the “Spatial detail” to 2. Higher value will let closer pixels to be grouped together.
3. Set “Minimum segment size in pixels” to 20. If a group of pixels is less than this value, that groups will be merged to its neighbor.
4. Uncheck “Show Segment Boundaries Only”
5. Next

V. Training Samples Manager

1. 1. Delete the default settings by right click the class and “Remove Class”
2. Right click the top category (right above the first class just deleted), “Add New Class”. It can be a main category or a sub category. Name it, give it a value which is just like a part number/model number, and set a color.
3. If a main category is created in step 2, you can still right click it and create subcategories under it. Same procedure as step 2. For example, in this step, a layer “Water” is created
4. Click “Water” class, zoom to water on the “Preview_Segmented” map, click the polygon tool, draw polygon over the pixels that only have water. Do it several times in different locations. Double click to finish the polygon.
5. Select the samples with same class, and “Collapse” them together.
6. Do the same thing on other class and objects.
7. Next

VI. Train

1. Set the “Classifier” to Support Vector Machine. This can handle large images and it’s less susceptible to discrepancies in training samples.
2. Set the “Maximum Number of Samples per Class” to 0. Maximum number of samples to use for defining each class. If want to use all samples, set it to 0.
3. Run

VII. Merge Classes

You can choose to merge the old class (on the left) to new class (right). New class is either itself or its higher category.

VIII. Reclassifier

1. Check and uncheck the layers so the newest merged layer and the initial RGB layer are checked, all else are unchecked.
2. Click Preview_Reclass, click “Raster Layer” tab, and click Swipe in the Compare section.
3. Drag the line to see the difference and inconsistency between two layers.
4. If there’s any error, click “Reclassify within a region” in the ‘Reclassifier, set the New class to which the object belongs to, and draw polygons around it.
5. Run, and Finish.