1 October 1994 Generation of neural-network decision surfaces: a pattern-classification example
William A. Schmidt, Jon P. Davis
Author Affiliations +
Abstract
The problem of classifying regions in an IR cloud scene is considered using the conventional matched filter, the perceptron, and a class of neural networks known as the multilayered perception (MLP). The purpose is to determine if neural network technology is applicable for a remote-sensing application. Another goal is to illustrate graphically how the MLP generates the decision surfaces. Examples are given to show that the MLP is better than other approaches if the application requires nonlinear decision surfaces for classification. A three-dimensional classification problem is given to illustrate the MLP hidden-layer operation. The resulting nonlinear decision surface, which achieves 100% correct classification, is displayed. A seven-dimensional classification problem is given to further illustrate the need for a nonlinear decision surface. Finally, an 128 x 128-pixel IR cloud-scene example with remote objects provides a more practical indication of the benefits of MLP for point-object classification. In addition, these examples provide a tutorial on how the internal operations of the MLP produce the required decision surfaces. The results provide a rationale for consideration of the MLP for other remote-sensing applications.
William A. Schmidt and Jon P. Davis "Generation of neural-network decision surfaces: a pattern-classification example," Optical Engineering 33(10), (1 October 1994). https://doi.org/10.1117/12.179387
Published: 1 October 1994
Lens.org Logo
CITATIONS
Cited by 1 scholarly publication.
Advertisement
Advertisement
RIGHTS & PERMISSIONS
Get copyright permission  Get copyright permission on Copyright Marketplace
KEYWORDS
Clouds

Image classification

Neural networks

Sensors

Remote sensing

Optical filters

Signal processing

Back to Top