Welcome to the Harris Geospatial product documentation center. Here you will find reference guides, help documents, and product libraries.


Harris Geospatial / Docs Center / ENVI API / Classification Framework / Classification Framework Glossary

Classification Framework Glossary

Classification Framework Glossary

Term

Definition

Activation function

The algorithm that performs the classification. Specifically, it is the algorithm that maps the attribute values in the examples to the activation values of the classes.

Activation value

A value given to each class for a given example. It could be the probability of an example belonging to the class. Typically, the class with the largest activation value is assigned as the class for that example.

Attribute

A unique feature of the data to be classified. For imagery, attributes include things like spectral mean, elevation, texture, and shape.

Class

Something that you want to identify in your data. For imagery, classes may include water, impervious surfaces, or vegetation.

Class Value

An integer that identifies a specific class.

Classifier

An instance of a classification algorithm with parameters set to specific values. A classifier takes data with a specified set of attributes and assigns each data item to one of a specified set of classes. Classifiers are of a certain type such as Softmax Regression and Support Vector Machine.

Example

A one-dimensional array (vector) of attribute values for one data item. For pixel-based classification, it is a vector of attribute values for each pixel. See Prepare Data for Classification for details.

Loss

The difference between truth values and the values predicted by the classifier. It is also called cost and is similar to residual, misclosure, distance, and delta in other scientific disciplines.

Loss Function

An algorithm that computes the loss of a classifier during training. It can be computed once per training iteration to inform the trainer how to update the classifier. The goal is to minimize the loss, so the loss function is also called a minimization function, cost function, or objective function.

Parameters

Classifier variables that can be changed internally. Classifier parameters are used in the activation function and updated as part of the training process. One example is the Theta parameter in the Softmax Regression classifier.

Properties

Classifier and trainer variables that are not changed internally. Examples include Learning rate (Gradient Descent trainer) and Lambda (Softmax Regression classifier).

Softmax regression

A generalization of logistic regression used for multi-class classification where the classes are mutually exclusive. See Softmax Regression Background for details.

Trainer Using data with a known classification, a trainer sets parameters on a classifier so that it can be used to classify new data. It is designed to minimize a loss function and to update the classifier parameters, typically over a number of iterations. Training can occur in a number of different ways, resulting in a number of different trainers.

Training data

Examples that you are certain belong to a given class. Training data are used to classify examples of unknown identity into a known class. In ENVI, you can use regions of interest (ROIs) to collect training data.



© 2017 Exelis Visual Information Solutions, Inc. |  Legal
My Account    |    Buy    |    Contact Us