Underwater targets classification using local wavelet acoustic pattern and Multi-Layer Perceptron neural network optimized by modified Whale Optimization Algorithm

Considering heterogeneities and difficulties in the classification of underwater passive targets, this paper proposes the use of Local Wavelet Acoustic Pattern (LWAP) and Multi-Layer Perceptron (MLP) neural networks to design a real-time and accurate underwater targets classifier. To train the MLP classifier, first, the Whale Optimization Algorithm (WOA) is improved and then applied to optimize the parameters of the designed classifier. For this purpose, different mathematical functions are employed for improving the exploitation and inspection capacity of the modified Whale Optimization Algorithm (mWOA). To evaluate the functioning of the proposed optimization algorithm and designed classifier, 23 benchmark test functions are used and an experimental underwater passive dataset is developed, respectively. To assess the accuracy of the classification, the speed of the convergence, and entrapment in local minima, the findings are compared with the results of five newly proposed meta-heuristic algorithms Biogeography-based Optimizer (BBO), Gray Wolf Optimizer (GWO), Salp Swarm Algorithm (SSA), Group Method of Data Handling (GMDH), and Harris Hawks Optimization (HHO), as well as classic WOA. The findings show that the modified optimizer and the designed classifier using mWOA significantly outperform the other benchmark classifiers.

Automated summary

This study proposes the use of Local Wavelet Acoustic Pattern and Multi-Layer Perceptron neural networks to design an underwater target classifier and improves the Whale Optimization Algorithm for parameter optimization. The results show that the modified optimizer and designed classifier outperform other benchmark classifiers in terms of performance.