7. Classifying An Image

"Classifying an image" means grouping/clustering pixels of similar wavelength into larger groups. Grouping areas with similar reflectance can help analyze patterns and quantify land cover types.

In an unsupervised classification, the computer groups similar reflectance. Often the cluster feature can be identified and named.

In a supervised classification, known features are selected by you and used to classify the image.

Below are directions for performing an unsupervised classification, followed by directions for performing a supervised classification.Further instructions are on the MultiSpec tutorial.

Unsupervised Classification:

1. Open MultiSpec
2. Open a file or subset file. the assignment of bands to channels does not matter.
3. Choose Processor, and then cluster.
4. Under Algorithm, check ISODATA.
5. In the next pop up: "Set ISODATA Cluster Specifications", you may want to reduce the number of clusters your first time through to 4. Click OK.
6. In the box: "Cluster Stats", select the choice: Do not save. Under "Write Cluster Reort/Map To:" check Cluster mask file and image window overlay. Leave Text window checked.
7. Click OK. Choose where you want to save it and click save. Wait for program to classify.
8. Optional: Study MultiSpec Text Output. The data values for each category are there. To compare cluster image locations to non-clustered image locations and to help identify cluster feature, keep original cluster image open, but click on the small red “0: in the lower left corner and choose no layers. This “toggles” the view to pre cluster layer.
9. Keep this image open so you can use it for a reference.
10. Go to File, Open image. Choose the file you just saved. it will end in: clMask.tif Click OK in the next pop up.You will see a color key in the left menu for each cluster in your .clMask image.
11. On the image with the color key, shift click on a color on the image. (you may want to magnify the image some).The curser turns to an eye. This will cause the cluster areas blink and help you rename the clusters
12. To change cluster names and colors: Double click the cluster color chip to change the color; double click the name to the right of the color chip to rename the each cluster as “vegetation”, “Streets/city" or "impervious surface”, “Water”; and to group classes together, go to drop down menu “classes”, then “group classes”. Click and drag one group into another group and rename. If you need help in naming the groups, go back to the other open images with the red 0 to identify surface types or go to Google Earth and zoom in.
13. Go to file, save as: "save image as GeoTIFF as and rename slightly, or use the shortcut Command shift s or close file and save when prompted to.
14. To find area for each class, study MultiSpec Text Output. I recommend saving the Text Output Box for further review.

Supervised Classification (-may need update: refer to MultiSpec tutorials)

1. Open MultiSpec
2. Open a file.
3. Click File. Select New Project. If New Project is grayed out, quit MultiSpec and reopen. Click OK on next window.
4. Highlight box “new” in Class box if not highlighted.
5. Draw a small rectangle around a homologous area.
6. Select “add to list”.
7. Enter a new class name that represents your training area, such as “vegetation”, click OK . If your area is small or if you want to include more sample fields in that lable, choose the same lable, draw another box, choose add. This adds another training field to that category.
8. Select “New” in the Class Field Class Field box.
9. Repeat steps 4-7 as needed.
10. Select Processor. Choose Classify. de-select check mark near Image Selection. choose ok and then choose update.
11. Find/Click on MultiSpec Text Output where the TRAINING CLASS PERFORMANCE is shown. the OVERALL CLASS PERFORMANCE should be close to 100% (90+ is pretty good).
12. choose Processor, choose classify.
13. Uncheck: the training (re-substitution).
14. Add check marks to: the Image Selection, the Window Image Overlay , Disk file, and Create Probability Results File. Click OK.
15. Leave name as is and choose save and save again in next popup.
16. Choose File, save project. keep the name but add something such as sc for supervised classification.
17. View the text output box to check percentages. Total percent classified should be close to 100%