V

Visualization

AToMS: Analysis Tool of Microscopic Structures. Context of modeling morphology in SEM images Images generated through microscopy techniques like scanning electron microscopy allow us to observe materials at over 5000 times magnification. Therefore, they are a critical tool in ‘seeing’ the microstructure of elements and composites: this leads us to directly infer the underlying physics of interaction between the grains by changes in their shapes, propagation of cracks in hard materials, etc. The analysis of such images requires a high level of expertise since we often have to distinguish between features produced through experimental shortcomings ( for example, damage in certain parts of the images due to experimental issues) and real underlying physics (for example, change in grain shape due to pressure and temperature induced mechanisms). At the same time, this work is extremely tedious, because of the volume of data produced. We propose an automated analysis of SEM images based on Segment Anything Model and a user interface to help experts. This is a proof of concept, more work still needs to do to evaluate the performances. Implementation was done by Benjamin Pianet and Logan Monnier under supervision of Amrita Chakraborti and Vincent Itier.