Ship Detection in Aerial Images Using YOLOv10

Real-time object detection pipeline for maritime monitoring

  • Built a YOLOv10 pipeline to detect ships in aerial images under scale and background variability.
  • Curated dataset, designed a custom DataLoader, and performed hyperparameter tuning for accuracy/latency trade-offs.
  • Trained with standard augmentations and evaluated using mAP, precision–recall, and confusion analysis.
  • Profiled inference for real-time feasibility with post-processing (NMS) configuration.

Repo: GitHub Dataset: Kaggle

Detected Ship Prediction Outputs

Highlights

  • Frameworks: PyTorch, YOLOv10
  • Data: Curated aerial imagery; augmentations; optional tiling for small objects
  • Training: LR scheduling, mixed precision, validation hooks
  • Metrics: mAP@0.5 / mAP@0.5:0.95, precision, recall; PR curves

References