Object-Based Classification

 Object-Based Classification in ArcGIS Pro

Kendrick Wittmer

4/8/2025

    The objective of this assignment was to follow a tutorial published by Esri regarding object-based classification and create my own set of instructions for the process. The following steps provide instructions for the completion of supervised object-based classification and the final product from following these steps can be found at the end of the post. 

1.     Extract the Spectral Bands

a.     After importing the desired datasets into ArcPro, first you will use the Extract Bands function to extract only the desired bands. In this case we are classifying impervious vs pervious surfaces, so extracting NIR (4), Red (1), and Blue (3) helps identify natural vs human made surfaces.

b.     Type 4 1 3 in the box for combination, and specify best match for missing band action to select what will occur if a listed band is unavailable

c.     Create and name a new layer with the extracted bands

2.     Configure the Classification Wizard

a.     Open the classification wizard under image classification

b.     Set supervised as the classification method and object based as the type with default schema

c.     Select desired output location

d.     Click next

3.     Segment the Image

a.     In this section you will group similar pixels into segments

b.     Set spectral detail to 8, spatial detail to 2, minimum segment size in pixels to 20, and uncheck show segment boundaries.

c.     This will create a preview of your segmented image in which similar pixels are grouped together to ease the classifying process

4.     Classify the Image

a.     In this next portion, you will set up the process of classification

b.     First you must create training samples

c.     Remove the default classes

d.     Add in desired classes with corresponding colors

e.     For this instance, we have impervious with a value of 20 and pervious with a value of 40. This value will be attributed to all subclasses

f.      Now we will add subclasses

g.     All impervious subclasses will begin with a 2, and pervious with 4

h.     Color accordingly

i.      For each subclass, draw polygons in areas containing the corresponding surface for classification (i.e. draw polygons on roofs for “Roof” subclasses)

j.      After completing training samples for a subclass, collapse the samples

k.     Draw around 5-6 training samples per subclass

l.      Hit next

m.   In the Train tab, set classifier to support vector machine with 0 being the maximum number of samples per class

n.     Click Run to generate a preview

o.     If you are dissatisfied with results, try adding some more training samples

p.     Click next if satisfied

q.     Appropriately name output and run the classification

r.      Click next in the image classification wizard

s.      In the New Class column, specify either pervious or impervious for the corresponding subclasses

t.      This will generate a preview with only two classes, pervious or impervious

u.     Click Next

5.     Reclassify

a.     If you notice any errors within your classification, you can reclassify specific areas

b.     In the classification wizard, click Reclassify within a region

c.     Draw a polygon on the error

d.     Set current class to any and choose the correct class for a region for New Class

e.     In the classification wizard, type your desired name for Final Classified Dataset and run to view your final product

Final Product

    The final supervised classification of pervious and impervious surfaces turned out fairly well. There are a couple spots where water was confused for a man-made object and roofs were classified as impervious, however adding a few more well-defined training samples may help with this.