Page123

Pokhara is the capital of the Nepalese district of Kaski

During the following exercises we will apply different classification methods. The aim is to understand the principles of image classification and compare the results produced by different methods.

Classification is a very useful tool used to retrieve information for planning, control and cartographic updates. It is a comparatively cheap and easy way to obtain information on land cover, land use and land changes, especially in remote or inaccessible areas.

However, even in well-known parts of the world, satellite images are part of our daily lives. Satellite images and classification maps are also used in cartography. Just think of how accurately forest boundaries are drawn in topographical maps. They are taken from satellite images and their image processing products.

Any digital classification, be it supervised or unsupervised, is only a basis for further adaptation. There is no way of producing an accurate and useful classification without manual input. The information processing carried out by the human brain allows the application of more complex procedures than a computer programme. There are so many complex relations between different types of surfaces which are not covered by spectral or geometrical differences.

A relatively simple classification method is Unsupervised Classification. All pixels in an image are grouped into a specified number of classes based on the similarity of their greyscale values.

Open the following images in LEOWorks:

anapurna_landsat_2000_SE_band_2.tif
anapurna_landsat_2000_SE_band_3.tif
anapurna_landsat_2000_SE_band_4.tif

Select Multivariate Analysis>Unsupervised Classification and select the images:

anapurna_landsat_2000_SE_band_2.tif
anapurna_landsat_2000_SE_band_3.tif
anapurna_landsat_2000_SE_band_4.tif

Type 7 for Nr. of Classes and 10 for Nr. of Iterations. The procedure is executed for the given number of iterations, or until a condition is met. Save the new image as Pokhara_432_2000_unsupervised in TIFF format in the folder Annapurna.
 
 
LEOWorks randomly assigns colours to the classes. The next step, therefore, is to change the colours so that they are more similar to their 'real-life' colour. Double click on the LUT bar at the bottom of the image frame. Select a colour in the pop-up menu and change the colour by sliding the RGB slides. For vegetation choose a green colour, for bare soil a beige colour and so on. When all the colours have been changed, save the image again.

Have a look at the classified image and compare it with the near-natural colour image and the infrared false-colour image.

Try other unsupervised classifications using different spectral bands, for example bands 1,4,7 or 1,2,3,4,5,7. Compare the various classification results.

Page123