Parameterizing neural network models to improve land classification performance - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 89

TABLE 7.1 List of some commonly-used neural network types in machine learning.

Brief description a No Network types

Multilayer feed-forward perceptron
networks

They comprise an input-hidden-output layered structure, and the input signal
propagates through the network in a forward direction. They are the most
widely used networks.

1

Radial basis function networks They use radial basis functions to replace the sigmoidal hidden layer transfer

function in multilayer perceptrons, thus transferring the design of a neural
network as a curve-fitting problem in a high-dimensional space.

2

Self-organizing networks They use a supervised or an unsupervised learning method to transform an

input signal pattern of arbitrary dimension into a lower dimensional (usually
one or two dimensional) space with topological information preserved as
much as possible.

3

Adaptive resonance theory (ART)
networks

They provide the functionality for creating and using a supervised or un
unsupervised neural network based on the Adaptive Resonance Theory.

4

Recurrent network (e.g., Hopfield
network)

Contrary to feed-forward networks, recurrent neural networks use
bi-directional data flow and propagate data from later processing stages
to earlier stages.

5

Modular neural networks (e.g.,
Committee of machine)

They use several small networks that cooperate or compete to solve
problems.

6

Stochastic neural networks (e.g.,
Boltzmann machine)

This type of networks introduces random variations, often viewed as a form
of statistical sampling, into the networks.

7

Dynamic neural networks They not only deal with non-linear multivariate behavior, but also include

learning of time-dependent behavior.

8

Neuro-fuzzy networks They are a fuzzy inference system in the body which introduces the

processes such as fuzzification, inference, aggregation and defuzzification
into a neural network.

9

a Detailed discussions on these neural network types are given elsewhere (e.g., Bishop,1995; Haykin, 1999; Duda, Hart and Stork, 2001).

Neuron or
Processing
Element (PE)
Weighted link

Output Layer

Input Layer

Hidden Layers

FIGURE 7.1 A fully-connected multilayer perceptron (MLP) neutral network with a 4 × 5 × 4 × 2 structure. This is a feed-
forward architecture. Data flow starts from the neurons in the input layer and moves along weighted links to neurons in the
hidden layers for processing. Each hidden or output neuron contains a linear discriminant function that combines information
from all neurons in the preceding layers. The output layer is a complex function of inputs and internal network transformations.