Data-drive-based Machine Learning and Singular Spectrum Analysis to identify Optical Patterns in Harsh Environments

This paper proposes an enhancement of a singular spectrum analysis (SSA) technique based on Hilbert Matrix (SSA-H) for autonomous navigation systems in a harsh environment. SSA is a technique that has caught the attention of different fields, and is used to extract trends and patterns in the time series domain. In this work, SSA-H is used to transform raw optical signals into meaningful information to build machine learning (ML) models. A technical vision system (TVS) with laser scanning for depth measurements is presented. According to the implementation of the TVS outdoors, some particular issues, such as interference radiation detected by its photodiode, can affect the system’s performance in determining depth measurements. This research extracts the laser beam patterns in a real environment to create ML models to solve the problem and address some of the technical challenges in a real environment. The main contribution of this work is designing an ML framework for recognizing the laser beam of the TVS based on SSA-H. Furthermore, a comparative analysis of ML techniques to discriminate sunlight interference was studied and compared with different configurations of SSA known in the literature. According to the results, the use of SSA-H can enhance the ML models such as Ensemble Learning (EL), Feed-Forward Neural Networks (FNN), Support Vector Machines (SVM), Naive Bayes (NB) in conjunction with autocorrelation function coefficients (ACC) as features.