A NOVEL MACHINE LEARNING METHOD TO DETECT DIFFERENT PHASES OF CONCRETE USING X-RAY MICROTOMOGRAPHIC IMAGES

Author(s): Sanat Kumar Saha, Subhasis Pradhan, Sudhirkumar V. Barai
Paper category: Proceedings
Book title: IV International Conference Progress of Recycling in the Built Environment
Editor(s): Isabel M. Martins, Carina Ulsen, Yury Villagran
ISBN:
e-ISBN: 978-2-35158-208-4
Publisher: RILEM Publications SARL
Publication year: 2018
Pages: 130-137
Total Pages: 08
Language : English

Abstract: This paper presents a novel machine learning based digital image processing technique to
segregate the three phases of concrete; such as voids, aggregates, and mortar. In this context,
the 8-bit images of concrete specimens obtained from X-ray microtomography (XRT) are
operated. Voids can be isolated assuredly owing to a clear distinction in the grey values of air
void from that of aggregates and mortar. For this the threshold grey value of void is
determined by observing the variation in grey value profile near the void edges. However, this
thresholding segmentation technique does not suffice to isolate the aggregates from mortar
because of the overlap of their grey values. Hence, a machine learning based technique is
developed in order to address this problem. This technique uses three parameters – (a) grey
value, (b) radial distance from the centre of the sample and (c) a weighted mean of the grey
values of neighbourhood pixels to determine a logistic regression based decision boundary
between the aggregate and void pixels. The present model efficiently and accurately detects
the three major phases of concrete.

Online publication : 2018
Publication type : full_text
Public price (Euros) : 0.00