ABSTRACT
This chapter shows how neural networks can be used as powerful tools for quantitative extraction of relevant information from high resolution transmission electron microscopy (HRTEM) images. It analyses different data preprocessing and modelling strategies. Atomic configurations in crystals can be obtained using HRTEM. Since images are available in digital form using charge-coupled device cameras it is possible to apply pattern recognition techniques to HRTEM image interpretation. There is a great number of parameters that affect the HRTEM image generated by a computer simulation, crystal structure, sample thickness, defocus value, incident beam direction, incident beam energy and crystal tilt. Feature extraction starts from raw data to construct a more compact representation in which the relevant information is retained. In doing this, there are three different alternatives. In order to reduce the dimensionality of the input space, only a few coefficients should be used.
