A method of underwater target detection via nonnegative matrix factorization

Reverberation is the unavoidable interference for an active sonar. It makes difficult to detect an underwater target. The theory of low-rank and sparse matrix decomposition provides a new idea for underwater target detection. Inspired by this theory, we propose a method via non-negative matrix factorization (NMF). The method first extracts the low-rank structure and then separates targets from reverberation through iterations. Simulation is performed to verify its feasibility and to evaluate with the ROC curve.