Identifying shark species using an AlexNet CNN model

Identifying marine life is vital for maintaining biodiversity and environmental health, but current methods are hindered by the need for time-consuming, labor-intensive manual observations. This research introduces a convolutional neural network (CNN) model designed to accurately identify shark species. Our primary goal was to overcome challenges posed by limited datasets through transfer learning and pre- trained models. We hypothesized that an AlexNet CNN model would achieve superior accuracy in classifying shark species, compared to other CNN architectures, conventional algorithms, and custom neural networks, especially within the constraints of a limited dataset. AlexNet’s deep convolutional layers and hierarchical learning capabilities were expected to enable effective feature extraction and learning from the limited data. Despite the challenge of working with a smaller dataset—only 100 images per species versus the recommended 5,000 samples per class—AlexNet’s ability to capture spatial hierarchies and patterns led to enhanced performance. Our investigation involved comprehensive experimentation and comparative analysis to validate this hypothesis, offering insights into optimal shark species classification in resource- constrained scenarios. The model was trained on a Kaggle dataset containing 1,400 images across 14 shark species. We employed AlexNet as a feature extractor, with fine-tuning steps to adapt the network to this dataset. Experimental results showed that our model, termed "SharkNet," achieved a 93% accuracy on the test set, surpassing conventional methods. This promising performance in distinguishing shark species could significantly aid marine biologists and ocean conservationists in monitoring and protecting these species.