Automatic event detection for noisy hydrophone data using relevance features

In this paper, a new context-aware method for detecting events in noisy hydrophone data is proposed. The method transforms first the 1D hydrophone data into a 2D relevance map. A dynamic context-aware relevance features set is then proposed extracted from the normalized relevancy map. Feature classification is finally performed using a least-squares support vector machine (LS-SVM). The method shows event detection sensitivity in excess of 97% for rare events such as whale calls from original noisy hydrophone recordings from the NEPTUNE Canada project, with more than 94% specificity and 95% overall accuracy. With relatively less parameters to adjust and high accuracy, the proposed method is useful for automated long-term monitoring of a wide variety of marine mammals and human related activities from hydrophone data.