Underwater Object Classification: Sonar-based Rock vs. Mine Detection

Abstract

The identification of rocks and minerals has been significantly enhanced with the advancement of SONAR technology, which relies on specific parameters to determine whether a detected object is a rock, a mine, or another entity. Machine learning has gained widespread attention across technology-driven industries due to its advancements in predictive analytics. The primary objective of this study is to develop an effective predictive model using machine learning algorithms to classify SONAR-detected objects as either rocks, mines, or other structures. This research presents a case study that implements a machine learning-based approach for categorizing rocks and minerals using a large, spatially complex SONAR dataset. The model achieved an accuracy of 83.17% with an AUC of 0.92. Further optimization using feature selection within the random forest algorithm improved the accuracy to 90%. Performance was assessed using standard classifiers such as SVM and random forest, alongside evaluation metrics like accuracy and sensitivity. The findings highlight the significant role of machine learning in enhancing underwater resource detection, with the potential for further advancements in the future. To ensure a balanced representation of both classes, the dataset was split into training and testing sets using a stratified approach.

A Logistic Regression, SVM, and KNN models were trained on the training data and evaluated for accuracy on both training and testing datasets. The model achieved notable accuracy scores, demonstrating its effectiveness in distinguishing between rock and mine signals.