Deep Granulometry

This repository contains the data related to the paper "Deep Granulometry: Image based estimation of concrete aggregate size distributions using deep learning" where a deep learning based method is proposed for the image based determination of concrete aggregate grading curves (cf. video).

Watch the video

More specifically, the data set consists of images showing concrete aggregate particles and reference data of the particle size distribution (grading curves) associated to each image. It is distinguished between the CoarseAggregateData and the FineAggregateData.    

Coarse Aggregate Data

The coarse data consists of aggregate samples with different particles sizes ranging from 0.1 mm to 32 mm. The grading curves are designed by linearly interpolation between a very fine and a very coarse distribution for three variants with maximum grain sizes of 8 mm, 16 mm, and 32 mm, respectively. For each variant, we designed eleven grading curves, resulting in a total number 33, which are shown in the figure below. For each sample, we acquired 50 images with a GSD of 0.125 mm, resulting in a data set of 1650 images in total. Example images for a subset of the grading curves of this data set are shown in the following figure.

Example images and grading curves of the coarse data set

   

Fine Aggregate Data

Similar to the previous data set, the fine data set contains grading curves for the fine fraction of concrete aggregate of 0 to 2 mm with a GSD of 28.5 $\mu$m. We defined two base distributions of different shapes for the upper and lower bound, respectively, resulting in two interpolated grading curve sets (Set A and Set B). In total, 1700 images of 34 different particle size distributions were acquired. Example images of the data set and the corresponding grading curves are shown in the figure below. Example images and grading curves of the finedata set

Related publications:

If you make use of the proposed data, please cite one of the publication listed below.

  • Coenen, M., Beyer, D., and Haist, M., 2023: Granulometry Transformer: Image-based Granulometry of Concrete Aggregate for an automated Concrete Production Control. In: Proceedings of the European Conference on Computing in Construction (EC3).

  • Coenen, M., Beyer, D., Ponick, A., Heipke, C. and Haist, M., 2023: Deep Granulometry: Image based estimation of concrete aggregate size distributions using deep learning. To be published.

Source Code

Source code for CNN based prediction of grading curves using the data set can be found on github under following Link.

BibTex: