Unsupervised vs Supervised Classification in Remote Sensing
Previously, we’ve explored digital image classification techniques like unsupervised classification, supervised classification and object-based.
Also, we’ve gone into great detail how to do object-based image classification.
Now, it only makes sense to practice supervised and unsupervised classification with some examples.
Supervised Classification in Remote Sensing
When you run a supervised classification, you typically go through the following 3 steps:
- Select training areas
- Generate signature file
In ArcGIS, here’s how to perform a supervised classification:
Step 1 Enable Image Analysis Toolbar
First, you have to enable the Image Analysis Toolbar (Windows > Image Analysis). After enabling it, the image analysis window opens in ArcMap.
Step 2 Select training areas
In order to add training samples, you can click the draw polygon icon. Now, you will have to draw polygons where you know the land cover class.
For example, draw a polygon for an urban area. And continue drawing urban areas representative of the entire image, not just a single area. After you have done a few, select all of your urban polygons and merge them into a single class.
For clarity, you can rename this training set as “urban”. Once you’re finished, begin creating training sets for your other classes.
Step 3 Generate signature file
At this point, you should have training samples for each class. In addition, you have merged each class and renamed them accordingly.
It’s time to create a signature file by clicking the “create a signature file” icon. If you’d like to revise or add additional samples, you can open it again at a later time.
Step 4 Classify
The ArcGIS image classification toolbar gives several options for classification including: maximum likelihood, iso cluster, class probability and principal components.
Each option has its own advantages but it’s best to test each one for yourself. As input, you will need your signature file which has the training samples. In other words, this generates a classified image with the classes you developed in your training set.
You may need a bit of trial and error with the signature files. Again, you can edit your signature file and rerun the classification until you are happy with the results.
Unsupervised Classification in Remote Sensing
Unsupervised classification is different because it does not provide sample classes.
First, the user identifies how many classes to generate and which bands to use. Next, the software then clusters pixels into the set number of classes. Finally, the user then identifies the land cover classes.
Unsupervised Classification Steps:
- Generate clusters
- Assign classes
Step 1 Activate Spatial Analyst Extension
First, you have to activate the spatial analyst extension in ArcGIS (customize>extensions>spatial analyst).
Step 2 Generate clusters
In this unsupervised classification example, we use iso-clusters (spatial analysis tools>multivariate>iso clusters).
INPUT: The image you want to classifiy.
NUMBER OF CLASSES: The number of classes you want to generate during the unsupervised classification. For example, if you are working with multispectral red, green, blue and NIR bands, then the number here will be 40 (4 classes x 10).
MINIMUM CLASS SIZE: This is the number of pixels to make a unique class.
When you click OK, it will be form clusters/classes based on your input parameters. But you still need identify which land cover classes each cluster belongs to.
Step 3 Assign classes
The last step is to identify each class from the iso-clusters output. In general, it helps to select colors for each class. For example, set water as blue for each class. After setting each one of your classes, we can merge the classes by using the reclassify tool.
There will be some manual classification if classes appear in 2 land cover classes. For example, if vegetation was mistakenly classified as water (perhaps algae in the water), then user will have to manually edit the polygon.
In most cases, it helps to convert the raster to vector and use the editing toolbar. You can split polygons to help properly identify them.
Classifying Images with Supervised and Unsupervised Methods
This sums up some of the basics for unsupervised classification in remote sensing.
We generated unknown classes (isodata) using iso clusters. Next, the user identified each cluster with land cover classes.
Some manual editing may be necessary if there is confusion between classes.
Put these steps to practice and generate some land cover of your very own.