The ocean, covering 70% of Earth’s surface, plays a critical role in maintaining ecological balance and biodiversity, supporting millions of marine species. However, analyzing underwater environments poses significant challenges due to poor visibility, color distortion, and an imbalance in available data about certain marine species. These factors hinder accurate identification and monitoring of marine life, making it difficult to address conservation needs effectively. We have developed two advanced methods that are specifically designed for underwater environments. The first, Object-GCN, merges YOLOv5 with GCNs to enhance the accuracy of marine species detection. By using BERT, the second method constructs a label correlation matrix that guides spatial attention to key image areas. Their collaboration allows for the precise classification of multiple species, even in complex underwater scenes. Through experiments on a custom underwater dataset, our models have been shown to perform better than traditional methods with higher accuracy and mAP scores. Their classification solution is strong due to semantic correlations, spatial attention, and advanced designs. This work supports future research and enhances efforts to monitor and preserve marine ecosystems.