The urgent necessity to identify and mitigate the damaging effects of Crown-of-Thorns Starfish (COTS) on coral reefs has spurred research for enhanced detection techniques. Manual methods, susceptible to monotony and inaccuracies, highlight the demand for more efficient approaches to safeguard these vital marine ecosystems. This study delves into the relatively unexplored application of deep learning in COTS detection, presenting a thorough comparison of recent approaches such as YOLOv5, YOLOv5 with TensorFlow Lite, and CNN with attention models. Leveraging the unique CSIRO dataset tailored for COTS detection through deep learning, the analysis reveals that CNN with attention models, incorporating well-known pre-trained models VGG19 and MobileNetv2, outperforms others in precision, recall, and F1 score. Despite achieving accurate object detection, the study highlights the undisclosed features influencing decisions in CNN with attention models. This research paves the way for marine ecology exploration, utilizing deep learning for rapid and precise Crown-of-Thorns Starfish detection, with significant implications for ecological monitoring and conservation efforts.