skip navigation


Register for Latest Research

Stay Informed
Register with NCJRS to receive NCJRS's biweekly e-newsletter JUSTINFO and additional periodic emails from NCJRS and the NCJRS federal sponsors that highlight the latest research published or sponsored by the Office of Justice Programs.

NCJRS Abstract

The document referenced below is part of the NCJRS Virtual Library collection. To conduct further searches of the collection, visit the Virtual Library. See the Obtain Documents page for direction on how to access resources online, via mail, through interlibrary loans, or in a local library.


NCJ Number: 134463 Find in a Library
Title: Neural Networks in Forensic Science
Journal: Journal of Forensic Sciences  Volume:37  Issue:1  Dated:(January 1992)  Pages:252-264
Author(s): C Kingston
Date Published: 1992
Page Count: 13
Type: Report (Technical)
Format: Article
Language: English
Country: United States of America
Annotation: Neural networks have been developed to understand and mimic the performance of the human brain, particularly with problems of forensic science interest.
Abstract: Humans are good at pattern recognition; the question is how good neural networks are at it. Simulation experiments with a neural network known as the Hopfield net indicate that it may be valuable in the storage of toolmark patterns, including bullet striation patterns, and in the subsequent retrieval of the matching pattern using another mark by the same tool for input. The back propagation network (BPN), another type of neural network, is useful in applications similar to those for which standard statistical methods of pattern classification can be used. This network is an appropriate approach to the matching of general component patterns, such as gas chromatograms of gasoline, or pyrolysis patterns from materials of forensic science interest, such as paint. The BPN may provide better results than statistical methods, but it is necessary to try both the Hopfield net and the BPN to determine which is best for a given situation. Both networks generally require long learning times, which makes it prudent to try competing statistical methods initially to see if they provide adequate discrimination. Technical descriptions of both networks, including mathematical equations, are provided. 6 references, 3 tables, and 6 figures (Author abstract modified)
Main Term(s): Forensic sciences; Toolmark identification
Index Term(s): Bullet hole identification; Chromatography
To cite this abstract, use the following link:

*A link to the full-text document is provided whenever possible. For documents not available online, a link to the publisher's website is provided. Tell us how you use the NCJRS Library and Abstracts Database - send us your feedback.