Objective: This study uses the improved capabilities of two powerful Convolutional Neural Net- work models for real-time object detection, YOLOv5 and Faster R-CNN, to improve the accuracy and efficiency of underwater fish detection in a variety of complicated aquatic situations. Methodology: The research involves a comparative analysis of YOLOv5 and Faster R-CNN models. These models are evaluated based on their detection speed, accuracy, and reliability in underwater settings. YOLOv5 is known for its speed and lightweight architecture, while Faster R-CNN emphasizes detection accuracy through region proposal mechanisms. Various underwater scenes, including those with blurred images, are used to assess the performance of both models. Results: The findings indicate that YOLOv5 excels in scenarios requiring real-time detection due to its speed. Conversely, Faster R-CNN demonstrates superior precision in complex image contexts, achieving an accuracy of 89.8%. This level of accuracy ensures reliable and swift fish detection even in blurred underwater scenes. Conclusion: This study provides valuable insights into the strengths and weaknesses of YOLOv5 and Faster R-CNN in fish detection tasks. It contributes to the advancement of marine biology studies, sustainable fishing practices, and the preservation of aquatic ecosystems by offering robust tools for accurate and efficient fish detection.