opensar-insight/vessel-detection-model

Vessel Detection Model

Inference models for detecting vessels (including dark/non-cooperative vessels not broadcasting AIS) in Sentinel-1 SAR data, developed as part of the OpenSAR Insight project. See the organization card for full project background, funding, and consortium details.

• Codebase: https://github.com/ESA-PhiLab/OpenSARInsight
• Project website: https://opensarinsightweb.web.uah.es/
• Training dataset: opensar-insight/vessel-detection-dataset

Model description

This repository contains the inference models for vessel detection in range-compressed Sentinel-1 SAR data:

• Teacher model (unrestricted): yolo26s-pose_opensar_true_vessel_filtered_dataset_640px_unrestricted.pt — a YOLO26s-Pose model used as the main/teacher model.
• Student model (lightweight): yolo26n-pose_feature_kd_from_s_640_300ep_feat4_lightweight.pt — a YOLO26n-Pose model, distilled from the teacher via feature-based knowledge distillation, intended for faster/onboard deployment.

Both models were trained on the filtered, higher-quality subset of opensar-insight/vessel-detection-dataset. Vessels are detected as pose keypoints (bounding box plus orientation), supporting downstream estimation of heading in addition to location.

Files

File	Role
yolo26s-pose_opensar_true_vessel_filtered_dataset_640px_unrestricted.pt	Teacher model (YOLO26s-Pose, 640px)
yolo26n-pose_feature_kd_from_s_640_300ep_feat4_lightweight.pt	Distilled student model (YOLO26n-Pose, 640px, 300 epochs, feature KD)

Usage

These are Ultralytics YOLO-Pose checkpoints:

from ultralytics import YOLO

model = YOLO("yolo26s-pose_opensar_true_vessel_filtered_dataset_640px_unrestricted.pt")
results = model.predict("path/to/sar_patch.png", imgsz=640)

For dataset generation, training, and the full processing pipeline (Level-0 to range-compressed to inference), see backend/pipeline/dvd_use_case/ in the OpenSARInsight GitHub repository.

License

These model weights are distributed under AGPL-3.0, since the vessel-detection pipeline (backend/pipeline/dvd_use_case/) depends on Ultralytics YOLO, which is itself AGPL-3.0 licensed. Other components of the OpenSAR Insight codebase (dataset generation, RFI pipeline, geocoding, SARFI) are MIT-licensed — see the repository LICENSE for the full breakdown.

Citation

If you use this model, please cite the OpenSAR Insight project:

@misc{opensarinsight,
  title        = {OpenSAR Insight: ML-ready datasets and models for direct insight generation from raw SAR data},
  author       = {{Indra Space} and {INTA} and {Universidad de Alcal\'a de Henares}},
  howpublished = {\url{https://github.com/ESA-PhiLab/OpenSARInsight}},
  note         = {Funded by ESA \(\Phi\)-lab}
}