Oriented Object Detection and Classification of Offshore Tin Mining Vessels Using YOLOv11 in Bangka Belitung, Indonesia

Abstract

Offshore tin mining in Bangka Belitung, Indonesia, has created increasing challenges for coastal and marine spatial governance, particularly in distinguishing vessels associated with fishing and mining activities. Conventional monitoring approaches are often constrained by the spatial extent of offshore areas and limited access to vessel level information. This study investigates the capability of YOLOv11 based oriented object detection to identify and classify three vessel categories relevant to offshore tin mining activities, namely fishing boats, mining pontoons, and dredging vessels, using satellite basemap imagery. A dataset of 104 images containing 593 labeled vessel instances was collected from coastal areas of Bangka, Belitung, and Singkep Islands and annotated using oriented bounding boxes. Five YOLOv11 variants were fine-tuned using transfer learning and evaluated using precision, recall, mAP@50, and mAP@50–95. Among the tested models, YOLOv11l achieved the highest recall (0.971) and a mAP@50 of 0.974, while YOLOv11x obtained the highest precision (0.991) and mAP@50 (0.977). However, mAP@50–95 remained substantially lower across models, ranging from 0.616 to 0.650, indicating that precise object localization remains more challenging than detection at the IoU 0.50 threshold. The results demonstrate the feasibility of oriented object detection for distinguishing functionally different vessels in complex coastal environments. The approach provides a potential basis for satellite based spatial monitoring of offshore mining activities and can support marine spatial governance, although further validation using independent test data, temporal imagery, and field-based observations is required before operational deployment.