Models and methods for automated underwater object detection and recognition by means of fluorescence LIDAR

Fluorescence LIDAR systems have been shown to be valuable for detecting and recognizing underwater objects in several applications. In general, underwater recognition by fluorescence LIDAR systems is not trivial, because the laser-induced fluorescence signal originated from the detected object is obscured by its combination with water column backscattering contributions. Therefore, once an object is detected, water column effects included in the fluorescence return signal need to be compensated for prior proceeding with object recognition. In spite of the several studies carried out over the years, the literature concerning suitable signal processing methodologies for underwater object spectral recognition by means of fluorescence LIDAR systems is still at an early stage. The development of automated underwater object detection and recognition methodologies based on the spectral analysis of laser-induced fluorescence signals and water column effect compensation methodologies is hence required. The main goal of this thesis is to study and develop comprehensive signal models and robust methodologies for underwater object detection and recognition by means of fluorescence LIDAR systems. With¬in this goal, three main aims have been identified and achieved by the present work. First, an underwater fluorescence LIDAR simulator that is able to generate fluorescence backscattering LIDAR signals returned from a water column both in presence and in absence of an underwater object has been proposed in order to develop, test and improve underwater object recognition methodologies. Secondly, laser induced fluorescence remote spectroscopy has been exploited in combination with the use of signal processing methodologies developed ad hoc in order to enable automated underwater object recognition. Finally, an end-to-end analytical model for predicting the performance that may be obtained in underwater object spectral recognition applications in terms of several different performance metrics in a given operational scenario has been proposed. Specifically, this model has been designed by employing the proposed fluorescence simulator and the automated underwater object recognition methodologies. In the light of the results obtained, this thesis has shown a good ability of the designed simulator at reproducing immersion of underwater objects in different operational scenarios, by providing, thus, a valuable support to the development of underwater object detection and methodologies and to the performance prediction of a given fluorescence LIDAR system in a specific scenario. Moreover, the proposed underwater object recognition methodologies have allowed recognition of underwater objects in a fully automated fashion.