Skip to content
Licensed Unlicensed Requires Authentication Published by Oldenbourg Wissenschaftsverlag December 1, 2015

Large scale classification of spectral signatures

Klassifikation in einer großen Menge von spektralen Signaturen
  • Matthias Richter

    Matthias Richter is a PhD student at the Vision and Fusion Laboratory (IES), Institute of Anthropomatics and Robotics (IAR), Karlsruhe Institute of Technology. He works in close cooperation with his colleagues at the Fraunhofer Institute of Optronics, System Technologies and Image Exploitation (IOSB). His main research interests are the application of machine learning methods for industrial image processing and automatic visual inspection.

    Karlsruher Institut für Technologie (KIT), Institut für Anthropomatik und Robotik (IAR), Adennauerring 4, 76131 Karlsruhe, Germany, Tel: +49-721-6091-659

    EMAIL logo
    , Thomas Längle

    Thomas Längle is associate professor at the Karlsruhe Institute of Technology (KIT), Karlsruhe and the head of the business unit “Vision Based Inspection Systems” (SPR) at the Fraunhofer IOSB in Karlsruhe, Germany. His research interests include different aspects of image processing and real-time algorithms for inspection systems. He also offers lectures in computer science at the Karlsruhe Institute of Technology and initiates many possibilities for students to work on applied research.

    Fraunhofer-Institut für Optronik, Systemtechnik und Bildauswertung (IOSB), Fraunhoferstraße 1, 76131 Karlsruhe, Germany, Tel: +49-721-6091-212

    and Jürgen Beyerer

    Jürgen Beyerer is the director of the Fraunhofer Institute of Optronics, System Technologies and Image Exploitation (IOSB) and the head of the Vision and Fusion Laboratory (IES) at the Faculty of Informatics, Karlsruhe Institute of Technology (KIT). His main fields of research are: Automated visual inspection and image processing, fusion of heterogeneous information sources, information theory, system theory, statistical methods and metrology.

    Fraunhofer-Institut für Optronik, Systemtechnik und Bildauswertung (IOSB), Fraunhoferstraße 1, 76131 Karlsruhe, Germany, Tel: +49-721-6091-210

From the journal tm - Technisches Messen

Abstract

Hyperspectral sensors are becoming cheaper, faster and more readily available. Apart from industry applications, manufacturers push to bring compact devices into the end-consumer market. This development gives rise to many interesting applications such as the identification of counterfeit pharmaceutical products or the classification of food stuffs. These applications require precise models of the underlying classes. However, building these models from expert knowledge is not feasible. In this paper, we propose to use machine learning techniques to infer a model of many classes from an annotated dataset instead. We investigate the use of three popular methods: support vector machines, random forest classifiers and partial least squares. In contrast to similar approaches using support vector machines, we restrict ourselves to the linear formulation and train the classifiers by solving the primal, instead of dual optimization problem. Our experiments on a large dataset show that the support vector machine approach is superior to random forests and partial least squares in classification accuracy as well as training time.

Zusammenfassung

Hyperspektrale Sensoren werden billiger, schneller und breiter verfügbar. Neben dem Einsatz in Industrieanwendungen versuchen Hersteller solcher Sensoren kompakte Geräte zum Endverbraucher zu bringen. Diese Entwicklung ermöglicht viele interessante Anwendungen wie die Identifikation von gefälschten Medikamenten oder die Klassifikation von Lebensmitteln. Solche Anwendungen erfordern ein genaues Modell der zugrundeliegenden Klassen. Ein solches Modell aus Expertenwissen zusammenzustellen ist jedoch kaum durchführbar. In diesem Artikel schlagen wir stattdessen vor, ein Klassifikations-Modell mittels maschineller Lernverfahren aus einer großen, annotierten Datenbank abzuleiten. Dafür untersuchen wir drei populäre Methoden: Support vector machines, random forests und partial least squares. Im Gegensatz zu vergleichbaren Ansätzen, die ebenfalls support vector machines verwenden, beschränken wir uns auf die lineare Formulierung und lernen den Klassifikator indem wir das primale, anstatt das duale Optimierungsproblem lösen. Unsere Experimente auf einem großen Datensatz zeigen, dass der support vector machine Ansatz sowohl random forests, als auch partial least squares betreffend der Klassifikationsleistung und benötigtem Rechenaufwand zum Trainieren überlegen ist.

About the authors

Matthias Richter

Matthias Richter is a PhD student at the Vision and Fusion Laboratory (IES), Institute of Anthropomatics and Robotics (IAR), Karlsruhe Institute of Technology. He works in close cooperation with his colleagues at the Fraunhofer Institute of Optronics, System Technologies and Image Exploitation (IOSB). His main research interests are the application of machine learning methods for industrial image processing and automatic visual inspection.

Karlsruher Institut für Technologie (KIT), Institut für Anthropomatik und Robotik (IAR), Adennauerring 4, 76131 Karlsruhe, Germany, Tel: +49-721-6091-659

Thomas Längle

Thomas Längle is associate professor at the Karlsruhe Institute of Technology (KIT), Karlsruhe and the head of the business unit “Vision Based Inspection Systems” (SPR) at the Fraunhofer IOSB in Karlsruhe, Germany. His research interests include different aspects of image processing and real-time algorithms for inspection systems. He also offers lectures in computer science at the Karlsruhe Institute of Technology and initiates many possibilities for students to work on applied research.

Fraunhofer-Institut für Optronik, Systemtechnik und Bildauswertung (IOSB), Fraunhoferstraße 1, 76131 Karlsruhe, Germany, Tel: +49-721-6091-212

Jürgen Beyerer

Jürgen Beyerer is the director of the Fraunhofer Institute of Optronics, System Technologies and Image Exploitation (IOSB) and the head of the Vision and Fusion Laboratory (IES) at the Faculty of Informatics, Karlsruhe Institute of Technology (KIT). His main fields of research are: Automated visual inspection and image processing, fusion of heterogeneous information sources, information theory, system theory, statistical methods and metrology.

Fraunhofer-Institut für Optronik, Systemtechnik und Bildauswertung (IOSB), Fraunhoferstraße 1, 76131 Karlsruhe, Germany, Tel: +49-721-6091-210

Received: 2015-7-6
Revised: 2015-9-7
Accepted: 2015-9-7
Published Online: 2015-12-1
Published in Print: 2015-12-28

©2015 Walter de Gruyter Berlin/Boston

Downloaded on 28.3.2024 from https://www.degruyter.com/document/doi/10.1515/teme-2015-0040/html
Scroll to top button