DES-Pat: A novel DES pattern-based propeller recognition method using underwater acoustical sounds

Abstract

Purpose: This work aims to propose a new propeller recognition (propeller type classification) method by using a nonlinear pattern-based sound classification model with high prediction. This model consists of feature generation, feature selection, and classification phases. To test this model, five types of propellers are produced using a 3D printer. These propellers are categorized using number of wings. Material and Method: An experimental data collection environment was created and underwater sounds of these propellers were collected (https://github.com/orhanyaman/Propeller). To generate features from these sounds, a new nonlinear feature generation function is presented by using one of the substitution boxes (S-Box) of the data encryption standard (DES) block cipher. This S-Box determines the patterns. Therefore, this feature selector is called as DES-Pat. Results: The proposed DES-Pat generates a feature vector with a size of 512. By using Neighborhood Component Analysis (NCA), 150 the most valuable features were selected. The selected feature vector with a size of 150 was utilized as the input of the selected 12 shallow classifiers in 3 categories: Decision Tree, k Nearest Neighbor (KNN), and Support Vector Machine (SVM). Conclusion: The results show that these methods are very successful for underwater acoustical sound classification since Quadratic and Cubic SVMs reached 99.8% classification accuracies. (C) 2020 Elsevier Ltd. All rights reserved.