ArcGIS Tutorial: Performing classifications

The purpose of the classification is to assign each cell in the study area to a known class (supervised classification) or to a cluster (unsupervised classification). In both cases, the input to the classification is a signature file that contains multivariate statistics for each class or cluster. The result of each classification is to divide the study area in the map into several classes, some of which are known to correspond to the training samples, and others that are naturally generated to correspond to clusters defined by clustering. The process of dividing a position into several naturally generated classes corresponding to a cluster is also called a layer.

The Image Classification toolbar provides an integrated environment for implementing the multi-step workflow required to perform the classification.

　　Maximum Likelihood method

The cells in the same class are rarely homogeneous. This is especially the case when training samples are used to supervise classification. For example, if a broad-leaved tree in a dark environment has similar reflection characteristics to a conifer in a full-sun environment, the two types of trees will be attributed to the same class. For a training sample taken from the habitat where you want to find the bear, any location may contain sub-locations where the bear does not appear.

In, class A represents broadleaf trees, and class B represents conifers. How do I classify cells that fall into Class A and Class B? Should the cell be categorized as Class A or class B?

　　

The maximum likelihood classifier calculates the probability of a cell belonging to each class based on the given attribute value. The cell is assigned to the class with the greatest probability, which is what is called the "maximum likelihood method".

For the maximum likelihood classifier to run accurately, the following conditions should be assumed:

· The data for each band should be normally distributed.

· Each class should have a normal distribution in the multivariate attribute space.

· The prior probabilities of a class must be equal-that is, in the absence of attribute-value weights, all classes appear in the same probability.

If the various types of prior probabilities in the study area are unequal, the classes can be weighted. For example, if the Alaska satellite imagery is classified, forests and other vegetation types will have a higher priori probability than human housing. That is, the probability that the cell containing the housing is less than the cell containing some vegetation types. When a cell value falls into an overlapping part of the housing category and vegetation category, the location is more likely to contain vegetation rather than housing, which should be categorized accordingly.

This probability and weight logic is based on Bayesian decision rules. The actual probability values for each cell and category are determined by the various types of mean and covariance matrices (stored in the signature file).

To perform the classification, use the maximum likelihood classification tool. This tool requires the input bands in the Multiband raster and the single-band raster, as well as the corresponding signature files. It is important to determine the way in which classes or clusters are weighted. There are three ways to weight a class or cluster: Equality, a cell in a sample, or a file. If you select Equal, all classes are weighted with the same prior probabilities. If you select a cell in the sample, the prior probability is proportional to the number of cells in each class or cluster in the signature file. If you select a file, the prior file input control is activated, and a priori probability is read from the specified file. The culling score must be determined. The reject fraction determines the cell parts that are still not categorized due to the lowest correct allocation probability. The default value is 0.0, so each cell is categorized. You can create an optional confidence level. Finally, you must specify the name of the output raster.

　　Category probability

The category probability tool does not assign a cell to a category based on the highest probability of the output raster, but instead outputs a probability layer in one band per input class or cluster. Values at various locations within each band are used to store the probability that the cell belongs to a class or cluster based on the attributes in the original input band.

This feature can be useful in the following scenarios. Suppose you are classifying an image, one of which is forest and the other is wetlands. After running this tool, you find that a cell on the forest class output raster receives a probability of belonging to a forest class of 60%, while the probability of receiving a wetland class on a wetland output raster is 30%. At this point, you may want to classify the cell location as a wet forest instead of classifying it as a forest.

　　View Multivariate categories

　　Supervised classification

The following steps are used to perform the supervised classification:

1. 1. Identify the input bands.
2. 2. Generate a training sample based on the known location of the desired class.
3. 3. form a signature file.
4. 4. If necessary, view and edit the signature file.
5. 5. Run the classification.

　　Non-supervised classification

The following are the steps for performing an unsupervised classification:

1. Identify the input bands.

2. Define the number of clusters to create.

3. form a signature file.

4. If necessary, view and edit the signature file.

5. Run the classification.

Copyright NOTICE: This article for Bo Master original article, without Bo Master permission not reproduced.

ArcGIS Tutorial: Performing classifications