Underwater Sound Classification

Classifying sounds underwater gives us the possibility to learn more about the incoming sounds and their meaning. The algorithm developed classifies incoming sounds by removing the noise and comparing that sound's features with those of known identified sounds located in a database. The algorithm provides capabilities to recognize and save specific incoming sounds for instance localization of specific sea animals (endangered species), trespassers on sea, drowning/help seeking people as well as identification of current events (ship related events: starting engines, stopping engines, loading and unloading of cargo). Different comparison techniques are implemented into the algorithm to obtain results. Sensors which are based on sonar technology are able to pick up sounds at long range, unfortunately picking up sounds is a wide term. The problem occurs when incoming sound is seen as an collection of sounds which gives no real internal data to work with. Discovering new sounds or specific patterns given by a specified recognized sound is practically impossible when the incoming sound is a compressed chunk collection. The Sound classifier (SC) solves the compressed chunk collection problem and presents recognizable results. In these results the user is able to see the recognition of a certain sound. The technique used by the SC also provides a confidence level about the provided classification. Adding sound data to the SC also makes the system smarter in its classifications, as the internal workings are inspired by supervised machine learning. This paper describes a solution to classify sounds underwater using a classification algorithm. The solution regards identifying the incoming sounds and classifying the sounds in the data library. The used sound samples are real recordings of an underwater pomp working, motor engines running in water and a whale sound extraction in the sea. The illustrations shown further in the report are tests performed in Opensource editors.