Scholarlink

<p>Indonesia is an archipelago country which has more sea than land area that become a maritime trade route, so it is necessary to improve the safety of ship navigation, such as with ship detection for ship monitoring. As technology develops, remote sensing can be utilized in the maritime field using SAR data for ship detection. With the current advanced technology of computer vision, object detection can be progressively done using deep learning. The capture of these objects especially in the sea can be provided by SAR space-borne images. This research focuses on ship detection using variety of YOLO with 3 different datasets, such as model 1 using Sentinel-1 image with RGB composite which Sigma Nought VV polarization for red, Sigma Nought VH polarization for green, and Sigma Nought VV/VH polarization for blue. While Official-SSDD was used in model 2, and a combination of Sentinel-1 using Sigma Nought VH polarization and Official-SSDD was used in model 3. Then, ship detection model evaluated and validated using AIS (Automatic Identification System) data. The result shows Model 3 on YOLOv9 is the best model for ship detection in Surabaya area with mAP 43.90%, while for Banjarmasin area Model 1 on YOLOv4 is the best model with mAP 26.78%. YOLOv4 is the best algorithm with mAP 26.78% based on every model performance. Model 3 and 1 are the most suitable data for ship detecting in Surabaya and Banjarmasin area because this data contains Surabaya and Banjarmasin scene as YOLOv9 and YOLOv4 also improve model performance due to architectural improvement.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Ship detection based on variety of YOLO using multi temporal and polarization SAR images

  • 期刊论文

  • 发表日期：

    2025年1月8日

查看全文
  • Noorlaila Hayati,
  • Safira Naurin Qolbi,
  • I Ketut Aria Pria Utama,
  • Teguh Putranto,
  • Dendy Satrio,
  • Mokhamad Nur Cahyadi

摘要

Indonesia is an archipelago country which has more sea than land area that become a maritime trade route, so it is necessary to improve the safety of ship navigation, such as with ship detection for ship monitoring. As technology develops, remote sensing can be utilized in the maritime field using SAR data for ship detection. With the current advanced technology of computer vision, object detection can be progressively done using deep learning. The capture of these objects especially in the sea can be provided by SAR space-borne images. This research focuses on ship detection using variety of YOLO with 3 different datasets, such as model 1 using Sentinel-1 image with RGB composite which Sigma Nought VV polarization for red, Sigma Nought VH polarization for green, and Sigma Nought VV/VH polarization for blue. While Official-SSDD was used in model 2, and a combination of Sentinel-1 using Sigma Nought VH polarization and Official-SSDD was used in model 3. Then, ship detection model evaluated and validated using AIS (Automatic Identification System) data. The result shows Model 3 on YOLOv9 is the best model for ship detection in Surabaya area with mAP 43.90%, while for Banjarmasin area Model 1 on YOLOv4 is the best model with mAP 26.78%. YOLOv4 is the best algorithm with mAP 26.78% based on every model performance. Model 3 and 1 are the most suitable data for ship detecting in Surabaya and Banjarmasin area because this data contains Surabaya and Banjarmasin scene as YOLOv9 and YOLOv4 also improve model performance due to architectural improvement.