Sign up for our Newsletter

Publications

Pro118-2

Trainable WEKA Phase Segmentation on SEM/BSE Images of Slag Blended Cement Pastes



Author(s): Natalia M. Alderete, Yury A. Villagrán Zaccardi and De Belie Nele
Paper category: Proceedings
Book title: Proceedings of International Conferences (ICACMS) Advances in Construction Materials and systems Vol 2
Editor(s): Manu Santhanam, Ravindra Gettu, Radhakrishna G. Pillai, Sunitha K. Nayar
ISBN: 978-2-35158-194-0
e-ISBN: 978-2-35158-191-9
Publisher: RILEM Publications SARL
Publication year: 2017
Pages: 628-636
Total Pages: 9
Language : English


Abstract: Scanning electron microscopy with backscattered electrons (SEM/BSE) is a powerful technique that allows the visualization of polished cross sections with good reproducibility and level of detail. It is widely used to study the microstructure of cement-based materials and identify different phases in the cement paste. However, in some cases it is difficult to distinguish between some phases due to a similar grey level, as in the case of slag and portlandite. Then, X-ray elements mapping is necessary to help in the differentiation according to composition, but it can be quite time consuming and tedious with standard detectors. A machine learning tool, trainable WEKA segmentation (TWS), can be used to train a classifier by means of pixel grey values and segment the different phases automatically without any assistance of compositional mapping, transforming the problem into a pixel classification issue. The trained models can be improved by adjusting each class. The application of the model to the images results in a segmented image that can be used for quantification. In this paper TWS is applied for segmenting SEM/BSE images without the need of elements mapping. Slag blended cement pastes at different ages are studied. Results are compared with image analysis through elements mapping and selective dissolution. From this comparison, some information regarding the image density of the portlandite is derived.


Online publication : 2017
Publication type : full_text
Public price (Euros) : 0.00


>> You must be connected to view the paper. You can register for free if you are not a member