Ship detection from remote sensing images using deep learning

Ship detection from remote sensing imagery plays a critical role in many areas such as maritime traffic control, illegal fishing, oil discharge monitoring, maritime security and environmental protection. High-resolution satellite imagery makes ship detection more efficient by covering large areas, but the large data volume and real-time analysis requirements cause traditional methods to be inadequate. Deep learning-based approaches are of great importance to achieve high accuracy in ship detection. Deep learning models accelerate the analysis processes by minimising human errors and make the processing of large data sets more efficient. In this thesis, ship detection was carried out using Google Earth open-source data set with YOLO models. The dataset contains images of ships in different scales, shapes, orientations and weather conditions from coastal and offshore regions. In the study, all images were labelled in YOLO tagging format and analysed with YOLOv8 and YOLOv9 models. As a result of the model training process, the accuracy of the YOLOv8 model was 95.4% for mAP50 and 69.3% for mAP50-95. The YOLOv9 model, on the other hand, can detect more complex structures with higher accuracy thanks to the advanced GELAN network, and the mAP50 value is calculated as 96.0% and the mAP50-95 value is calculated as 73.7%. Model comparisons show that YOLOv9 can better distinguish small and close objects. The results show that YOLO-based deep learning models provide high accuracy in ship detection from remote sensing images. In particular, it is seen that YOLOv9 provides more successful results thanks to its advanced architecture. In this context, it is demonstrated that deep learning-based object detection methods can be effectively used in maritime applications and can significantly improve automatic ship detection processes. In addition, the development of these models with larger and more diverse data sets in the future will further advance remote sensing-based maritime analyses.