There are several difficulties in exploring, tracking, and identifying things in the vast and complex deep-sea environment. Visibility is extremely poor at depths where sunlight seldom reaches, making human involvement impractical. This calls for applying cutting-edge technology, including sonar and remotely operated vehicles (ROVs), to collect information. It is challenging to evaluate this data to categorize items, whether they be artificial waste, natural marine life, or navy warships. Machine learning (ML) is a branch of artificial intelligence that helps computers learn from data and make predictions or decisions independently, without needing to be specifically programmed for every task. To identify patterns and distinguish between diverse underwater items, such as marine animals, detritus, or submarines, machine learning models are trained on datasets in the field of marine object classification. This work aims to improve the accuracy and reliability of marine biodiversity identification using a tailored R-CNN model designed for underwater environments. By addressing challenges like image complexity and variability, our approach integrates advanced CNN architectures, data augmentation techniques, feature extraction methods, and model regularization strategies. By blending marine expertise with modern AI techniques, this study seeks to improve marine biodiversity identification and support important uses like maritime surveillance and ecosystem monitoring.