AppliedSensor Unveils Patent-Pending Data Processing Algorithm for Sensor Systems.

Business Editors

NEW ORLEANS--(BUSINESS WIRE)--June 25, 2001

Optimal Discriminative Projection (ODP) First to Automate

and Improve Performance in Specific Applications

Visit AppliedSensor at IFT, Booth # 4660

AppliedSensor, Inc., an emerging leader in the field of chemical sensor systems, has unveiled its patent-pending data processing algorithm - Optimal Discriminative Projection (ODP) - to solve the dimension reduction problem present in gas sensor systems.

This is achieved by reducing the dimensionality or number of variables that is associated with various sensor systems in multiple classes of data.

AppliedSensor's products use chemical sensors to identify volatile compounds or off-odors. This enhances and compliments quality assurance methods in the food, packaging, environmental control, and industrial service industries. Upon identifying a volatile off-odor, the presence of multiple sensors creates numerous variables. By decreasing the number of variables in a given application by automating the algorithm, ODP offers a gas sensor system the best possible performance during processing.

"ODP offers an optimal classification producing an improved performance compared to other methods," said Tomas Eklov, AppliedSensor's senior R&D scientist. "ODP guarantees the optimal model in data processing in an area where many have had difficulties doing so."

ODP Improves Upon Other Methods

Compared to other existing data processing methods such as the Principal Component Analysis (PCA) and the Linear Discriminative Analysis (LDA), ODP offers enhancements that can accurately determine the quality of the processed data collected from a gas sensor system.

Unlike ODP, PCA does not take class information into account when determining dimension reduction. Therefore, in most cases, the best model will not be found when using PCA.

When addressing the dimension reduction problem, LDA focuses on separating the average distance between all class pairs, therefore, LDA isn't always certain to separate all classes from each other. Also, it is established that LDA does not avoid "overfitting", which is when the method uses overly optimistic models. This can occur if care is not taken in the application development phase. ODP has built-in features to avoid "overfitting", which will, to a large extent, avoid this problem.

ODP Benefits & Features

When tracking quality assurance in food applications, the first step is to examine the data from a multi-class problem such as different food products. By using the ODP method, the algorithm focuses on maximizing the minimal distance between any class pair; this will provide a large separation between food products ensuring a more accurate result.

"Providing a method that guarantees a completely automatic optimal model will certainly have an impact on the growing sensor market," said Tom Aiken, COO of AppliedSensor, Inc.

The patent-pending algorithm is available with the AppliedSensor measurement and analysis software, Senstool, and in the Matlab toolbox. The new algorithm includes the following features:

- Optimal classification that improves performance

- Automatic estimation making the process easier to track

- Built-in features to avoid "overfitting" - a term used to determine an overly
optimistic model

- Simple calculation which requires a simple system to furnish rapid
calculations

- Reliable performance to provide accurate predictions

- Operable with any sensor technology available today

About AppliedSensor Group

AppliedSensor Group was formed in 2000 to provide a range of chemical sensors for quality assurance. AppliedSensor, Inc. operates from Parsippany, N.J.; AppliedSensor Group is based in Linkoping, Sweden; and AppliedSensor GmbH is based in Reutlingen, Germany. Following the joining of Nordic Sensor Technologies (NST) and MoTech Sensorik, AppliedSensor Group was formed to offer a comprehensive portfolio of chemical sensor equipment with a range of industry-accepted technologies.

For more information, please visit the website at www.appliedsensor.com or contact Christopher Berardi at 212.840.5900x211, or by e-mail at [e-mail].