Improving a novel image-processing based method for automatic detection, extraction and characterization of marine mammal tonal calls by means of a segmentation process based on learning machines

Abstract

Marine mammal monitoring is becoming increasingly important for the detection, localization and identification of specific species of marine mammals. It permits the compilation of information that helps researchers to learn about marine mammal habits and communication skills. Furthermore, knowlegde of the presence of marine mammals in certain areas facilitates the protection of these animals. In this paper, it is proposed a new segmentation method of the marine mammal tonal calls based on the training of a learning machine. In particular, a MultiLayer Perceptron (MLP) has been trained by means of the Optimally Pruned Extreme Learning Machine (OP-ELM). The suggested approach is able to detect the information from a spectrogram in an automatic way. The proposed method has been tested with real signals and the results are very promising.