THE UNIVERISTY OF CALIFORNIA AT BERKELEY

IMAGE CLASSIFICATION USING XPACE

__________________________________________________________________________________________________

PURPOSE:

Alternatives of performing supervised image classification using “Xpace” in PCI

MATERIALS:

The Berkeley TM image (or the Fremont EO-1 ALI image at your own choice)

LANDCOVER/USE CLASSES you may want to identify:

High Density Residential, Low Density Residential, Commercial, Industrial

Road network, Forest, Shrub, Grassland, Baren, Water

STEP 1. Use “edit graphics” in “Imageworks” to select training areas and use “save graphics” in “Imageworks” to save each set of training areas for a particular class.

STEP 2. Use “CSG” in Xpace to generate class signature. Each time you can only generate a signature file for one class. Program “CSG” requires you to specify the “File name; Database Image Channels; MASK of Training Areas for a class; VALU (value) of the Class (i.e., the code for this particular class appearing in the final classification image); Threshold - number of standard deviations around the class means; BIAS - weighting factor applied to each class. Bias can be kept the same, usually 1. You can experiment with the different thresholds. Each run of “CSG” will generate one set of signature for one class.

STEP 3. Use “MLC” in Xpace to do the classification. You will need to input the following information in order to run MLC

FILE - Database File Name : file of your choice (e.g., berkeley.pix)

MAXL - Classifier: PARA/TIES/FULL:

PARA - Parallelepiped classifier

TIES - Parallelepiped but when boundary overlaps use mlc

FULL - MLC

DBS1 - DBase Class Signature Subset 1 : input selection of class signature segments

DBOC - Database Output Channel List } Specify the channel to hold the classification

results

PROBCHAN- Probability Channel List > Specify a list of channels to hold class

probabilities

MASK - Area Mask (Window or Bitmap) > Specify an area for classification

If classify the entire image, leave it empty

NULLCLAS- Null Class: YES/NO :YES

If this option is "YES" then a pixel is assigned to a class only if

it is within the Gaussian threshold specified for the class. If it

is not within any threshold, it is assigned to the NULL (0) class.

If the option is "NO" then the thresholds are ignored and every

pixel will be assigned to the most probable class (i.e., nearest

class based on Mahalanobis distance).

REPORT - Report Mode: TERM/OFF/filename :TERM

Can save the classification report into a file of your choice (e.g., mlc1.res)

Would you like to compare the class means in each band?

Are you curious how well the classifier does to your training samples?

Do you have a set of independent samples for each class and would like to know how accurately they are classified?

To answer the above questions, you need to use three programs in

STEP 4. Use “CSR” in Xpace to output class means. Fill the following parameters

FILE - Database File Name :

DSIG - Database Class Signature Segment>

REPORT - Report Mode: TERM/OFF/filename :

STEP 5. Use “MAP” in Xpace to create a set of samples for accuracy assessment.

MAP Database Bitmap to Channel Encoding

This program converts bitmap masks into image brightness values

For classification, those bitmaps (masks) should be encoded as they were assigned in CSG

FILE - Database File Name : File of your choice

DBIB - Database Input Bitmap } MASK

VALU - Grey-level Value List > Corresponding value for each mask

DBOC - Database Output Channel List > A new channel to hold the converted mask

STEP 6. Use “MLR” in Xpace to generate a summary of classification accuracy

FILE - Database File Name : MISSING

UNITS - Reporting Units :HECTARES

DBIC - Database Input Channel List > Channel holding the classification

DBSA - Database Sub-Areas Channel > Channel created by “MAP”

DBS1 - DBase Class Signature Subset 1 } List of signature segments used in

MLC

MATRIX - Show Confusion Matrix: YES/NO :

It is helpful to analyze the confusion matrix so create it.

MASK - Area Mask (Window or Bitmap) >

Use a mask only if you are interested in a report on the accuracy for

the masked area.

REPORT - Report Mode: TERM/OFF/filename :TERM

Questions:

Given the image - Berkeley.pix and the following classification scheme:

High Density Residential, Low Density Residential, Commercial, Industrial

Road network, Forest, Shrub, Grassland, Baren, Water

Conduct the classification. Think about the following questions.

1. How did you select training samples?

2. How are the training samples determined? Did you use any map or airphoto to assist you?

3. How many samples in each class did you use in your classification?

4. Provide a report of the classification accuracy using independent samples for verification.

Summarize the accuracy.

5. What are your suggestions for improving your classification results?

Are there any need to adjust the classification scheme?

Is it possible to improve training?

6. Suppose you have some undesirable classification results and you would like to edit the results, what procedure would you propose? Can you do some automatic adjustment?

Options:

7. Examine the ISOCLUS or KCLUS programs in Xpace and see if you can produce a classification using these programs. Read the help to find what are their differences.

8. Can you edit the clustering algorithm and get the classification map containing the 10 classes as specified above?

9. Can you find a program that converts the classification results into a vector GIS format?