Identification of marine species is crucial to the ecological conservation and biodiversity monitoring yet under water imagery is subject to low contrast and sparse labeled data. This paper introduces a deep learning-based classification system of six marine creatures (crabs, jellyfish, seahorses, sharks, star-fish and sea turtles). To enhance our models, we optimized three state-of-the-art convolutional neural networks (ResNet50, InceptionV3, and EfficientNetB0) with transfer learning and added large data augmentation to balance classes and enhance robustness. Experimental results demonstrate that EfficientNetB0 achieved the highest performance (nearly 99% accuracy) among the models, with superior precision, recall, and F1-scores on the validation data. We also developed a simple graphical user interface to enable real-time species identification. These results show that modern deep learning techniques can reliably classify marine species from challenging underwater images, offering a practical tool to support marine ecology research and conservation efforts.