TPU-Enabled SAR Image Processing in Satellites for Maritime Surveillance

This project develops an advanced marine image processing system for maritime surveillance applications. State-of-the-art object detection models, YOLOv11 and YOLOv10, were implemented and fine-tuned to accurately identify key maritime entities with the following classes: boat, buoy, jetski, life_saving_appliances, and person. The models were trained and evaluated on a specialized dataset obtained from Roboflow, allowing detailed performance comparison across metrics. A user-friendly web application was built using HTML, Flask, and CSS, incorporating local user registration and secure login features. Authenticated users can seamlessly upload marine images and receive immediate detection results, complete with bounding boxes and class labels overlaid on the images. The system offers an effective solution for enhanced maritime monitoring, supporting improved safety measures, search and rescue operations, and environmental protection efforts.

Keywords: Maritime surveillance, marine image processing, object detection, YOLOv11, YOLOv10, deep learning, boat detection, buoy detection, jetski detection, life-saving appliances, person detection, Roboflow dataset, web application, Flask, user authentication, image upload, bounding box visualization