Präsentation: Forschungsergebnisse von Benjamin Seppke am 01.12.2015 um 14 Uhr
1. Dezember 2015, von David Mosteller
Title:
A Concurrent Skeleton-based Approach for the Characterization of Wood Fibers with Sub-pixel Precision For Fiber Board Production
Abstract:
The automatic derivation of fiber properties by means of mass data (image) analysis is a current and interesting research area. Skeleton-based approaches have already proven to be adequate in paper pulp quality control. However, the derivation for thermo mechanical pulp (TMP) used for the production of medium dense fiberboards (MDF) is, due to the wide range of fiber lengths and fiber morphology, far more challenging. Currently, trained experts are performing the quality assessment visually and haptically on small samples. Since this cannot replace an objective measurement, we propose a fast skeleton based approach, which is able to derive a variety of fiber characteristics at sub-pixel precision from images of fibers. The algorithm is supposed to work automatically under a laboratory environment for fiber samples, and makes use of concurrency to reduce the processing time and the memory needed. In practice, one measurement cycle is based on approx. 1700 single images with usually more than 1,000,000 automatically measured fibers. Comparisons of the automatically measured fibers with manual microscopic measurements yield in a high reliability and quality of the results.
Zur Darstellung der Stärken dieser Repräsentation werden des Weiteren zwei Anwendungen zur Straßenverfolgung sowie zur Dacherkennung für myanmarische Militärgebäude im Bereich der Verifikationsaufgaben in der Friedensforschung präsentiert. Für eine Beschleunigung der Verarbeitung werden zudem innerhalb der Repräsentation und den Applikationen verschiedene Techniken zur Parallelisierung verwendet.