How do you classify a raster or satellite image so that land cover can be easily identified and possibly measured? For this assignment we are to develop land cover classes of the Black Water Wildlife Refuge using remotely sensed data. Since we already know what many of the spectral reflectance signatures are for vegetation, developed areas and many other classes of cover, we know that it should be fairly straightforward to come up with a land cover classification for the Refuge.
Most satellite images are viewed in true color- or how we see them with our eyes. For image classification it's better to use a false color image so that we can better distinguish certain types of trees, water or vegetation. Now we can better pick training samples that will group all pixel values with the same value into the category we assign to it. After I've selected enough samples to come up with a proper classification, I run the tool to create an output.
An initial map was created and then compared to the original image. In order to achieve a higher accuracy, more training samples were added or amended, resulting in a final land cover classification image.
(Original satellite image in false color
with training samples selected) (Final image with land cover classes)
(The area calculation of each classified
Application and Reflection
Image classification is a great tool for being able to readily see different types of land cover and also when you want to differentiate between different types of vegetation. It's also useful for looking at land cover change from several points in time.
A farmer wants to estimate how much seed he will need to plant his crops in the spring and has enlisted my help to calculate the area of his 6 fields.
High resolution image of the target land area.
The image should be processed in false color so that a more accurate measurement can be taken of the fields. Taking proper training samples will make sure you have a homogenous classification or if the image is good quality, polygons may be drawn around the fields. Once the image is classified, you can create a new field in the attribute table of the output image and create a calculation for area based on the cell size.
ESRI Training Module on Unsupervised and Supervised Image Classification
Using an area in Queenstown, NZ a raster image is classified into 25 classes using unsupervised classification. The spectral classes are then grouped into 4 classes.
This module teaches how to perform both supervised and unsupervised classification.
(Raster image of Queenstown, NZ) (Iso Unsupervised classification with 25 classes)
(Classifying Imagery using ArcGIS)