The main challenge from a computational perception point of view was the detection of an hexagonal mesh of nano-structures from an EBM image:

After a thresholding operation and an extraction of connected components, the hexagonal mesh can be fitted by detecting the dominant inter-blob distance. The difficulty stems out of the missing nano-structures and from a slight projective transformation (homography) due to the non-orthogonality of the imaged object with respect to the electron beam. The solution implements an EM-inspired solution, iteratively estimating the homography and the mesh-coordinates of the nano-structures. The thresholded nanostructures can be seen below (green for the disk-shaped blobs, red otherwise, blue for the mesh neighbourhing relations).

The fitted mesh wil look like the following:

Our results on the processing of electron-beam microscope images has been used in the following papers:

  • El Gmili, Y., Bonanno, P. L., Sundaram, S., Li, X., Puybaret, R., Patriarche, G., Pradalier, C., Decobert, J., Voss, P. L., Salvestrini, J.-P., & others. (2017). Mask effect in nano-selective-area-growth by MOCVD on thickness enhancement, indium incorporation, and emission of InGaN nanostructures on AlN-buffered Si (111) substrates. Optical Materials Express, 7(2), 376–385.
  • Sundaram, S., Li, X., El Gmili, Y., Bonanno, P. L., Puybaret, R., Pradalier, C., Pantzas, K., Patriarche, G., Voss, P. L., Salvestrini, J.-P., & others. (2016). Single-crystal nanopyramidal BGaN by nanoselective area growth on AlN/Si (111) and GaN templates. Nanotechnology, 27(11), 115602.