A Multi-Faceted Approach for Underwater Sonar Rock vs Mine Classification Using Machine Techniques

A significant challenge of underwater object classification distinguishing sonar rocks from buried mines is crucial in maritime and naval security, as the enemy country has planned explosive mines in the ocean. Furthermore, there is a possibility of encountering rocks. This research is motivated by the urgent need for reliable and efficient underwater detection methods, which have a variety of applications in environmental studies and maritime security. Many classification methods, from traditional algorithms to modern deep learning models, have been used to identify the unique patterns in sonar data. This study enhances the accuracy of the existing system capable of classifying between mines and rocks beneath the submarine to protect human resources and marine life. Our comprehensive approach integrates data preprocessing, feature extraction, and several machine learning classification techniques such as KNN, NB, SVM, Decision Tree, Random Forest, and neural network by utilizing the Sonar Mines vs. Rocks data to classify the object. The outcome of this research shows that a high accuracy of 98% was achieved in the LDA grid approach followed by 94 % in SVM (rbf), and 88-91% in neural network and boosting techniques. This research work also provides a smart web interface allowing users to input sonar data and receive real-time classification results.