AI for High-productivity Digital Materials Characterization

While general-purpose AI chatbots and other consumer utilities are making a wider and wider impact on people’s individual lives, advanced fine-tuned AI solutions for scientific imaging continue to mature and deliver significant productivity to materials researchers. GeoDict integrates user-friendly AI solutions in the materials property characterization toolchain so that they are easily used to solve demanding image processing tasks. Here I describe three GeoDict artificial neural network AI solutions for image segmentation and image enhancement.

The tools we introduce here are found in three different modules, related to their function. The first I demonstrate and explain is the ImportGeo-Vol solution for greatly accelerating user workflows for segmenting scientific images into their constituent materials. After using simple painting tools to label representative areas, GeoDict will automatically train an AI model—either a convolutional neural network or a random forest model - to segment even the most challenging microstructures for you.

The second solution is the pair of FiberFind-AI and GrainFind-AI, which takes a material label and breaks it down into isolated fibers or grains, respectively. Following the distinct labeling of each fiber or grain instance, users then easily measure fiber length or grain size distribution and orientation, and use the statistical information for modeling and designing your desired material.

Finally, I will present GeoDict-AI, which is a tool for transforming low-quality scientific images laden with artifacts into high quality scientific images, more suitable for image processing. Importantly, this solution enables users to build and reuse their own neural network models for material labeling or image enhancement, such as contract stretching or denoising. The GeoDict workflow guides you in creating training material, conducting the training and evaluating the success of the training. Most significantly, scientists and engineers can immediately leverage AI for customized and fine-tuned solutions even without any previous AI experience.