This thesis discusses the application of deep learning to Coherent Fourier
Scatterometry data in order to quickly and reliably detect nanoparticles on
surfaces. An introduction to deep learning is followed by a review of the
experimental setup and used software. After that, results are presented of
classification accuracy tests on various datasets containing images obtained
from scatterometry scans. We show that a relatively simple convolutional
neural network can achieve an accuracy as high as 98% on a 200 image test
set. We compare this to the accuracy of a non-deep learning, clustering
based classification algorithm and conclude that deep learning is a more
suitable method for particle classification. Then, three methods of open set
recognition are applied. We show that it is possible to reject 80% of a fooling
dataset at the cost of rejecting 10% of the normal data. Finally, the results
are discussed and placed in the context of future work on this subject.