Identifying fish species accurately and efficiently is crucial for ecological research, sustainable fisheries management, and underwater exploration. Deep learning models trained on vast image and video datasets can automatically identify fish species with high accuracy, on-board fishing vessels, or from underwater cameras. This work proposes a deep learning-based approach for fish species detection using the lightweight and real-time object detector Tiny-YOLOv4-SPP. This work has been carried out on a publicly available FishPak dataset. The training and testing results with the Tiny-YOLOv4-SPP algorithm achieved a map value of 99.79% indicating it is useful for fish species detection. The results obtained with the proposed Tiny-YOLOv4-SPP are better than the Tiny-YOLOv4 and Tiny-YOLOv7. Accurate species identification leads to more precise population estimates, size structure analysis, and catch composition data. This work can be helpful in designing more targeted and effective management strategies for specific species, preventing overfishing of vulnerable populations, and promoting sustainable practices.