A Two-Layer Ensemble Deep Learning Model for Acoustic Seabed Sediment Classification

Seabed sediment information is widely utilized in marine exploration and management. Recent breakthroughs of sonar technology and data processing method have spurred the development of automated seabed sediment classification techniques. However, the variability and complexity of sonar data present a series of challenges that must be addressed before acoustic sediment classification technology can be practically applied. Traditional methods for classifying seabed sediments often struggle with the complexity and variability of acoustic data, leading to misclassification and instability in sediment type predictions. To address these challenges, we develop a robust ensemble framework that integrates several deep learning models, to leverage the strengths of each model. Experiments conducted on real sonar datasets demonstrate that the ensemble deep learning model outperforms any individual model and traditional ensemble learning model, achieving higher accuracy and better generalization. These results also highlight the potential of ensemble deep learning in tackling the intricacies of practical seabed sediment classification.