Simulation of microstructures and machine learning

TL;DR

This paper explores the use of synthetic images generated by stochastic models to overcome data scarcity in machine learning tasks like industrial defect detection and 3D crack segmentation.

Contribution

It proposes leveraging stochastic geometry models to generate diverse, labeled training data, addressing annotation challenges and data imbalance issues.

Findings

Synthetic images can effectively supplement real data for training.

Stochastic models capture structural variability and provide free ground truth.

The approach raises questions about fidelity requirements for real-world data.

Abstract

Machine learning offers attractive solutions to challenging image processing tasks. Tedious development and parametrization of algorithmic solutions can be replaced by training a convolutional neural network or a random forest with a high potential to generalize. However, machine learning methods rely on huge amounts of representative image data along with a ground truth, usually obtained by manual annotation. Thus, limited availability of training data is a critical bottleneck. We discuss two use cases: optical quality control in industrial production and segmenting crack structures in 3D images of concrete. For optical quality control, all defect types have to be trained but are typically not evenly represented in the training data. Additionally, manual annotation is costly and often inconsistent. It is nearly impossible in the second case: segmentation of crack systems in 3D images of concrete. Synthetic images, generated based on realizations of stochastic geometry models, offer an elegant way out. A wide variety of structure types can be generated. The within structure variation is naturally captured by the stochastic nature of the models and the ground truth is for free. Many new questions arise. In particular, which characteristics of the real image data have to be met to which degree of fidelity.