A hybrid recorded-synthetic sonar data set for validation of ASW classification algorithms

Modern anti-submarine warfare sonars are often designed with narrow beamwidths and wide frequency bandwidths in order to maximize spatial resolution and sonar performance. A known issue for high-resolution sonars in littoral environments, is the occurence of high false alarm rates. Increased false alarm rates increase the workload of sonar operators and also reduces the usefullness of automatic systems such as autonomous underwater vehicles, since their limited communication abilities hinder them from sharing large amounts of contacts. The false alarm rate may be reduced simply by increasing the threshold used in the detection process. However, this also reduces the probability of detecting actual targets. Automatic classification algorithms provide more sophisticated alternatives for false alarm reduction. The work presented here demonstrates an automatic classification algorithm on a data set collected in a littoral environment. The data set contains a large amount of false alarms, particularly close to the coast, but does not contain any submarine target detections. Synthetic submarine echoes are therefore added to the sonar data set. Six features are extracted from the hybrid synthetic-recorded data set. The features are fed into supervised machine learning schemes. The performance of each scheme is presented as receiver operating characteristic curves.