Stacking Ensemble Learning for Seabed Sediment Classification

Seabed sediment information is extensively applied in marine exploration and management. Recent advancements in sonar technology and machine learning have introduced automated techniques for sediment classification. However, it still faces a series of challenges before acoustic sediment classification technology can be practically applied, due to the variability and complexity of sonar data. Developing more refined and reliable data interpretation techniques is crucial for accurately identifying sediment types from noise-contaminated data. Various machine learning methods have progressively been introduced into the field of acoustic sediment classification to handle predictive tasks. This study explores the application of stacking ensemble learning method for classifying seabed sediment using sonar images. A meta-learner is designed for integrating predictions of a number of base learners. Comparative analysis with traditionally used models demonstrates the effectiveness of stacking architecture in distinguishing between different sonar images. The improvements in classification performance suggests promising applications of ensemble learning for accurate seabed sediment classification.