We study in this project the machine learning-based recognition of a laser beam measured and displayed in an intensity map. The characterization of the laser signal includes its spatially integrated intensity and its position on the map. We show firstly that unsupervised learning methods assist in predicting the intensity level, and then supervised k-means++ and CNN can determine the location of the laser spot. The percentage errors from various unsupervised learning methods are reported, and the relevant origins of such errors from either the model deficiency or the label pre-computation process are discussed. The excellent capability of k-means++ and CNN distinguishing real signals from impurities is also explained in this study.
The structure of this repository is very straight forward. The source codes are in the Codes folder, and the full report available as a PDF file.