Attractive time-variant orthogonal Schur-like representation for click-type signal recognition

Abstract

Analysis of click-type signals in the presence of noise with time-varying statistics is a challenging task, especially in low signal-to-noise ratio conditions. This well-known problem is commonly present in underwater passive acoustics applications. In this paper we present a novel solution for this dilemma as applied to marine mammal acoustics - a well-established basis for marine mammal study and protection. The adaptive orthogonal Schur-like algorithm is proposed to classify medium-frequency odontocete clicks. This technique is characterized by excellent convergence behaviour, very fast parametric tracking capability and robustness. The difficulty of recognition (classification) resides in the extraction of the signal's intrinsic information; i.e. extraction of an efficient signal signature. It is expected that the distances between the signatures within the class are minimal (small intra-class variance) and between the classes are maximal (high inter-class variance). This condition ensures a good recognition performance (separability of classes). The 2D signature proposed in this work and based on a selected set of the time-varying Schur coefficients assures this requirement. When compared to the classical Fourier approach, the proposed recognition method is four times as efficient for inter-class distances and twice as efficient for intra-class distances. The results of the recognition are given for sperm whale (Physeter macrocephalus) regular clicks and striped dolphin (Stenella coeruleoalba) nacchere clicks. They are very satisfactory and promising for other applications. The proposed technique can be easily implemented in real-time applications such as underwater acoustic monitoring systems.