<p>Marine scientists conduct research on specific fish species in their natural environments by remotely photographing and videotaping them. This advances their knowledge and predicting fish behavior to factors such as environmental degradation, loss of habitat, and fishing activity levels. This information is important for supports sustainable fisheries and for protecting the environment. But the huge number of images and videos that have been collected is labor-intensive and inefficient for a person to find useful information. Deep learning (DL) and Convolutional Neural Networks (CNN) are promising tools to solve this problem. DL and CNN can help marine scientists analyze large volumes of images and videos quickly and effectively, domain-specific insights that they can’t get with traditional monitoring methods. In this study, we propose a fish classification framework, termed the Fish Species Identification Algorithm (FSIA), which is based on fine-tuning the Inception v3 model. Rather than introducing a new algorithm, FSIA integrates tailored preprocessing, imbalance-aware training, hyperparameter optimization, and interpretability via Grad-CAM. Using a dataset of over 13,000 images of 31 distinct species of fish, FSIA achieved a top accuracy of 99.92%, demonstrating robustness and reliability. This study highlights the effectiveness of deep learning workflows for fish-type classification and shows how explainability and careful model design can support sustainable fisheries and marine conservation.</p>