An Adapative Machine Learning Approach For Autonomous Detection and Classification of Marine Life in Underwater Environments

Deep Learning technology, particularly in aquatic ecology, is crucial for recognizing marine organisms amidst the complexities of underwater environments. This study introduces a computer vision system using the 'Super Gradients' pre-trained model, a PyTorch-Based training library, for detecting and classifying marine species. Employing a newly developed Sea Animals Image Dataset with annotated images of fish, crabs, and starfish in brackish water with varying visibility, the 'Super Gradients' model demonstrated high accuracy in species classification. The annotated dataset enhances research by providing diverse training data. Key findings reveal the model's significant impact on detection and classification accuracy, the dataset's robustness, and the model's effectiveness in feature extraction. The integration of advanced deep learning and comprehensive datasets improves marine monitoring efficiency and reliability, addressing underwater environment challenges and offering future research insights. This study underscores the importance of automated systems in aquatic ecology, highlighting technological advancements and key concerns in the field.