Development of a fusion adaptive algorithm for marine debris detection within the post-Sandy restoration framework

Recognition of marine debris represent a difficult task due to the extreme variability of the marine environment, the possible targets, and the variable skill levels of human operators. The range of potential targets is much wider than similar fields of research such as mine hunting, localization of unexploded ordnance or pipeline detection. In order to address this additional complexity, an adaptive algorithm is being developing that appropriately responds to changes in the environment, and context. The preliminary step is to properly geometrically and radiometrically correct the collected data. Then, the core engine manages the fusion of a set of statistically- and physically-based algorithms, working at different levels (swath, beam, snippet, and pixel) and using both predictive modeling (that is, a high-frequency acoustic backscatter model) and phenomenological (e.g., digital image processing techniques) approaches. The expected outcome is the reduction of inter-algorithmic cross-correlation and, thus, the probability of false alarm. At this early stage, we provide a proof of concept showing outcomes from algorithms that dynamically adapt themselves to the depth and average backscatter level met in the surveyed environment, targeting marine debris (modeled as objects of about 1-m size). The project relies on a modular software library, called Matador (Marine Target Detection and Object Recognition)