Identification of Underwater Targets Based on Sparse Representation

We consider using sparse representations to identify underwater targets, since underwater acoustic signal have sparse characteristics. We consider the identification problem as one of the identifying among multiple linear regression models and believe that the new theory from sparse signal representation provides the key to solving this problem. Based on a sparse representation computed by ℓ
₁
-minimization, we propose a general classification algorithm for (hydroacoustic signal-based) targets identification. This new framework provides new insights into identifying two key issues in underwater targets: feature extraction and robustness of signal loss and noise interference. For feature extraction, we point out that feature extraction is no longer critical if the sparseness of the underwater acoustic signal is properly utilized. The critical is whether the number of features is large enough and whether the sparse representation is correctly computed. This framework can handle errors due to signal loss and noise interference uniformly by exploiting the fact that these errors are often sparse with respect to the standard (hydroacoustic signal) basis. Extensive experiments have been conducted based on a public underwater acoustic signal sampling set to verify the efficacy of the proposed algorithm and corroborate the above claims.