Author(s): N. Efthymiou; P. Rutschmann
Linked Author(s):
Keywords: No Keywords
Abstract: The bed material grain size distribution is usually obtained from time consuming and not so cheap sieve analysis methods. Non-intrusive image analysis of the bed layer may serve as an alternative. Recent image filtering techniques allow automated identification of particles and derivation of the corresponding data. The lack of a contrasting background, heterogeneity in grain lithology, texture and shape as well as the packing arrangement and shading effects make the automated identification of individual particles a difficult task. In an optimization process, a new two step procedure for image segmentation for large and small grain fractions respectively, is presented in the current study. The conversion of 2D segmented ellipses into 3D ellipsoids was improved by taking the flatness index of the grains into consideration. The quality of automatic segmentation and the grain size distribution from the numerical sieving process was checked by comparison with the manual segmentation and traditional sieve analysis methods respectively. The nonintrusive and automated procedure was developed to investigate temporal evolution of armored bed layers. The very encouraging results show satisfactory identification for grains having larger diameter than 1 mm, (for the pixel resolution used). Segmentation of smaller grains is knotty due to the above-mentioned problems. Finally with the use of a Plexiglas sheet slightly touching the water surface, we are able to obtain sharp images of the armored bed during ongoing tests.
Year: 2009