Supervised Classification in ERDAS Imagine

Classification is one of the very basic and important parts of Goespatial Technologies. Any satellite image will generally have 256 discrete values. Hence talking from layman’s point of view, every image will have around 256 classes. This implies that vegetation might be covered in 50 classes based on their DNs (As different type and stages of vegetation will have different spectral reflectance values).

Classification is basically a method which puts all these 50 values into 1 class of vegetation. So in general classification can be defined as the process of grouping all the pixels of an image into a specified number of classes.

Primarily there are two types of classification,

Unsupervised Classification: Discussed in unupervised Classification video in the blog.

Supervised Classification: This is type of classification that requires quite a bit of human intervention. Here the user will define something called signature set, which are primarily samples of the classes user is going to define. User will digitize a sample portion of a feature and then save it under s specific name. For example user will digitize a small potion of a waterbody and save it under the name water. Classifier will check the entire image and groups all the pixels with similar spectral properties of a specific spectral signature. This is the basics of Supervised Classification and as you can realise there is a need for the user to supervise the entire process. Resampling method selected will also affect the classification result to some extent.

Following is the video on Supervised Classification Using ERDAS Imagine.

Below is the video on classification if an image using ERDAS Imagine.

Notes and Tips:

• Accuracy of the classification only depends on the accuracy of the signature set. So take extra care while you define any signature
• If you are classifying the image into only a few classes say less than 5, may be unsupervised classification is the better solution
• If you feel there is mis-classification, you can edit the signature set and define a new class (or a class that is already existing, but you feel that the mis-classified portion should belong there) selecting the portion which you feel is mis-classified.  With this, you may end up having two classes of vegetation but likelihood of mis-classification will be addressed to some extent. While preparing map where you have more than one class for any feature, you can assign same colour to both the classes and keep only one class in the legend in the map