Base2Designs

Home ] Up ] Contents ] Search ]

MPhil Thesis
 

 

 

Download thesis

Title

UNIVERSITY OF SUSSEX METAL DETECTION USING NEURAL NETWORKS A THESIS SUBMITTED TO THE DEPARTMENT OF ENGINEERING FOR THE DEGREE, MASTER OF PHILOSOPHY BY STEVE FIELDING BRIGHTON, EAST SUSSEX SEPTEMBER, 1993

Abstract

The purpose of the work completed is to provide an improved method for differentiating industrial products contaminated with metal, from uncontaminated products. Graseby Goring Kerr, the sponsoring company, currently produce industrial metal detection equipment which provides reasonable detection capabilities, but the actual signal detection process is a simple amplitude thresholding system, and leaves a lot of scope for improvement. A neural network approach is used for signal classification; this was placed in context by reviewing other methods of pattern recognition. Neural networks provide the best approach to this problem because of their ability to cope with corrupted patterns, and the fact that no prior assumptions need to be made about the form of the signal. A WISARD net, which is a simple RAM based net, was used because of its relative simplicity and its ease of training. Choosing to use a particular pattern recognition technique cannot be done in isolation of the pattern type and the pre-processing that is to be used. A new fast optimal transform was used to pre-process the quantized sampled signal to provide a small number of coefficients for use as input to the WISARD net. Reduction in data size means that the net can be smaller and work faster, both important when the net is implemented in real time on cost sensitive hardware. Tests were carried out on stored data using programs running on a PC. This proved that improved detection of contaminated products is possible.

 

Send mail to sfielding@base2designs.com with questions or comments about this web site.
Last modified: 07/12/11