Designing a Sonar System with the Ability of Classifying Active and Passive Acoustic Targets Based on the Evolutionary Neural Network

Considering the importance of identifying and determining the nature of the sonar targets in marine battles, the purpose of this paper is to design a system with the ability to classify active and passive sonar targets using multi-layer perceptron neural networks (MLP NNs). Considering the defects of MLP NNs in dealing with real-world data, as well as low classification accuracy and low convergence rate, this paper proposes a new meta-parasitic algorithm called Chaotic Groups Particles Swarm Optimizer (CGPSO) to train an MLP NN. This algorithm explores the search space faster and better than normal particle swarm Optimizer (PSO) using chaotic and independent groups. To evaluate the designed system, a benchmark sonar dataset, a passive laboratory data set and an active dataset were developed. In order to have a comprehensive comparison, the designed system was compared with PSO, biogeography-based Optimizer (BBO) and Gray Wolf Optimizer (GWO) in terms of convergence rate, classification accuracy, and reliability. Results show that the designed system was more accurate than the best available classifier, by average 2.33%.