Context knowledge-aided sparse bayesian learning method for bathymetric sonar

Abstract

Sparse Bayesian learning has recently become successful in many compressed sensing problems. However, their performance critically relies on the appropriate tuning of numerous hyperparameters, because they directly control the statistical distribution and have a significant impact on the performance of the model being trained. Compared with manual tuning, grid search and random search, Bayesian optimization is a better hyperparameters intelligent optimization strategy. Nevertheless, traditional Bayesian optimization treats each problem as an independent black box. In this study, the hyperparameters tuning process is further improved because the contextual knowledge can be formulated into problem domain that is easier to optimize. For the bathymetric sonar, it is generally believed that the underwater acoustic environment changes slowly, which can be draw inspiration from the property and incorporated into the Bayesian optimization framework as additional prior information. This extension could yield efficient optimization without requiring too many iterative steps to converge. Simulation and tank experiment are conducted, and the results demonstrate that the proposed method has a lower computational cost in comparison with the conventional sparse Bayesian learning method.