Acoustic Detection and Recognition of Dolphins using Swarm Intelligence Neural Networks

Detecting and recognizing marine mammals have found serious attention recently, given the inhomogeneous underwater sound propagation environment. Meanwhile, the emphasis on applying machine learning and swarm-based intelligence algorithms has become stronger. This paper developed an accurate dolphin recognizer and proposed a multi-layer perceptron neural network (MLPNN). Despite many capabilities, MLPNNs suffer from severe built-in defects tackling real-world problems, including convergence rate, entrapment in local minima, and sensitivity to initialization. Therefore, the chimp optimization algorithm (ChOA) is first utilized to optimize the MLPNN parameters. Furthermore, the modified-chimp concept is introduced and applied to improve ChOA efficiency. Finally, a common benchmark dataset is used, followed by developing an experimental dolphin vocalization dataset to evaluate the performance of the designed model. The results are verified by a comparative study with Slime Mould Algorithm (SMA), Harris Hawks Optimization (HHO), Henry Gas Solubility Optimization (HGSO), and Kalman Filter (KF) approaches, as well as classic ChOA, based on recognition accuracy, convergence speed, and entrapment in local minima. The simulation results indicated that the weighted-chimp optimizer and the proposed recognizer outperform other benchmark recognition methods significantly.