Direktorat Jenderal Bea dan Cukai

<p>Smuggling violates Customs regulations through the illegal movement of goods across national borders. While Customs sea patrols are used to prevent smuggling, Indonesia's vast waters and unpredictable conditions make constant patrolling inefficient and risky. A more precise approach is necessary. This research explores how integrating satellite imagery with deep learning can improve Customs surveillance and monitoring capabilities. Using the CRISP-DM methodology, this research examines the potential of using YOLOv8 and YOLONAS with satellite imagery collected from Sentinel-2. YOLOv8 achieved a mAP50 of 0.465 and mAP50-95 of 0.269, while YOLONAS scored 0.409 and 0.235, respectively. Evaluation using 16 satellite images reveals that YOLONAS often misidentifies non-vessel objects, such as clouds and islands, as vessels, while YOLOv8 maintains higher accuracy without these errors. Additionally, YOLOv8's performance improves with smaller map scales, as evidenced by the increased confidence values for detected objects. The results of this study can be used to enhance Customs&#8217; ability to detect vessels in smuggling-prone areas and identify potential new smuggling routes, enabling more targeted and efficient sea patrols and thus preventing illegal activities along the Indonesian coastline.</p>