• data analysis
• detector characterisation
• gravitational waves
• image processing
• LIGO
• machine learning
• statistical signal processing
• Virgo

In December 2019, the LIGO Scientific Collaboration and the Virgo Collaboration have published the results of their first two joint observing runs (O1 and O2), describing the source properties of ten binary black hole (BBH) and one binary neutron star (BNS) events. Their validation has been made employing state of the art Data Analysis techniques. All of them rely, to some extent, on the assumption to know the statistical properties of the detector noise, from which the gravitational signals are extracted. Moreover, their performances are optimal, with respect to certain criteria, if the noise distribution is stationary and Gaussian. To this purpose, in this Thesis work we have studied several strategies aimed at the verification of the previous two hypotheses, and the characterisation of the detector noise. Once a specific noise feature, detrimental for gravitational wave searches, was found, we have proceeded to the investigation of its causes and some mitigation strategies. The techniques that we have implemented have been selected from many fields of research, like Digital Image Processing and state of the art Machine Learning. Two original contributions have been introduced. One consists of a new method for the identification of generic non-stationary noise, from the variations in the empirical distribution of the signal RMS value. The other is a wavelet-based, instantaneous causality statistic, specifically aimed at the study of transient noises. These aim to improve upon other existing strategies and have been applied for the investigation of specific noise issue in Advanced Virgo detector data.