Detection of snapping shrimp using machine learning

The marine environment consists of many different sound sources covering a wide frequency range. Accurately identifying and analysing these sound sources is difficult and time consuming. This is compounded by effects such as variable ambient noise, multi-pathing and multiple sources. One promising technique for analysing such complex data sets is machine learning. This has been successfully used in many other applications. In this work we will use it to detect snapping shrimp impulses. These are a dominant noise source in shallow tropical waters and ideal for testing new algorithms. The logistic regression method is used as the main algorithm. A snapping shrimp acoustics matrix (SSAM) is constructed from features such as the band energy ratio, frequency centroid, spectrum flatness, etc. It has been ensured that the extraction speed of the SSAM is sufficiently fast such that it is suitable for real time processing. A number of data sets for different locations covering a range of conditions will be analysed and compared.