| This thesis develops an algorithm that can determine if a laser is functioning correctly over a long period of time. A Fourier fit is created to model fringe profiles from a Fabry-Perot interferometer, and singular value decomposition is used to reduce noise in each signal. Levenberg-Marquardt gradient descent is performed to correctly locate the center of each image and to optimize each fit with respect to the spatial frequency. The Fourier fit is used to extract important information from each image to be used for separating the image types from one another. Principal component analysis is used to reduce the dimensionality of the data set and to plot a projection of the data using its first two principal components. It is determined that the image data are not linearly separable and require a non-linear support vector network to complete the classification of each image type. |
