Detecting objects on sea and ocean surfaces, particularly ships, is critical for various maritime applications, including traffic regulation, emergency response, environmental protection, and security. However, this task is challenging due to dynamic factors such as background variability, wave interference, and diverse lighting conditions. Satellites in low Earth orbit offer superior image clarity compared to geostationary counterparts, capturing dynamic features crucial for ship detection. Despite these advantages, challenges persist, including distinguishing ships from other structures like piers and managing high viewing angles that complicate object recognition. The deployment of deep learning techniques has revolutionized object detection, replacing traditional methods with models like R-CNN, Faster R-CNN, SSD, RefineDet, and YOLO. These algorithms enable precise localization and classification of various objects in diverse environments, enhancing security, autonomy, and surveillance systems. This study compares the performance of Faster R-CNN and YOLOv7 in ship detection from satellite images. Faster R-CNN, renowned for accuracy, employs a two-stage detection process, excelling in complex scenarios but at the cost of longer execution times. In contrast, YOLOv7, optimized for real-time applications, offers rapid detection by simultaneously predicting bounding boxes and class probabilities in a single pass. Evaluating these models involves analyzing metrics such as precision, recall, throughput, and execution speed, providing insights into their effectiveness across different operational requirements. The findings contribute to understanding the trade-offs between accuracy and speed in ship detection applications, guiding the selection and optimization of detection systems for maritime surveillance and security.