A Comprehensive study on Advanced Self-Supervised Hybrid ML-DL Framework for Underwater Sea Object Identification and Classification

Identification and classification of objects under water are critical tasks in exploring the underwater environment, surveillance activities, and navigations of autonomous underwater vehicles. However, issues like low visibility, light absorption, light scattering, and lack of datasets limit the capability of detection systems. While the latest developments in deep learning have made a difference, deep learning models depend on a large number of annotated datasets. One way of overcoming this challenge is through self-supervised learning since self-supervised machine learning enables the models to learn meaningful features without annotated training sets. This project introduces an innovative self-supervised machine learning technique that uses representation learning, feature enhancement, and classification in detecting objects under water. It will use unlabeled underwater images to extract features and classify them using supervised fine-tuning. Recent research findings reveal that hybrid methods make a huge difference in detection and generalize well across different environments [1][2][3].