The advancement of digital technology has increased the need for efficient and accurate object monitoring and detection systems to support national defense and security, particularly in the maritime domain. One of the persistent threats to maritime operations is the presence of submerged naval mines, which are difficult to detect due to low illumination, limited image contrast, and visual similarities between the target object and its surrounding environment. This study aims to develop a naval mine detection prototype based on You Only Look Once Version 8 (YOLOv8) using a synthetic underwater image dataset and to evaluate the performance of the model and its audio-based early warning system. The research methodology included data collection, image annotation using Roboflow, preprocessing, data augmentation, model training using Google Colaboratory, and real-time testing in a simulated environment using an external webcam and an artificial aquascape. The results showed that the model achieved a precision of 0.85, a recall of 0.80, an mAP50 of 0.87, an mAP50-95 of 0.43, and the highest F1-score of 0.84 at a confidence threshold of 0.382. The object was not detected under illumination levels of 0–100 lux but was successfully detected at illumination levels above 100 lux with an optimal distance of 45–60 cm. The system also successfully displayed bounding boxes and automatically activated an audio alert when a target was detected. These findings indicate that YOLOv8 has potential as a supporting technology for maritime defense and security early warning systems.