Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Maritime Small Object Detection from UAVs using Deep Learning with Altitude-Aware Dynamic Tiling

Sakib Ahmed, Oscar Pizarro

TL;DR

Maritime small-object detection from UAV imagery is challenged by low object-to-background pixel ratios at high altitudes. The authors propose altitude-aware dynamic tiling that combines altitude-dependent scaling with adaptive tiling inside the SAHI framework, paired with YOLOv5s, to maintain detection performance while reducing computation. On the SeaDronesSee dataset, the method yields a 38% relative improvement in $mAP_{50}^{small}$ over a no-tilting baseline and more than a threefold increase in inference speed compared with static tiling, illustrating a favorable accuracy–speed trade-off for real-time SAR operations. The approach offers a practical, altitude-responsive tiling strategy that complements existing detectors and can be extended to multi-modal data or alternative architectures for robust maritime UAV surveillance.

Abstract

Unmanned Aerial Vehicles (UAVs) are crucial in Search and Rescue (SAR) missions due to their ability to monitor vast maritime areas. However, small objects often remain difficult to detect from high altitudes due to low object-to-background pixel ratios. We propose an altitude-aware dynamic tiling method that scales and adaptively subdivides the image into tiles for enhanced small object detection. By integrating altitude-dependent scaling with an adaptive tiling factor, we reduce unnecessary computation while maintaining detection performance. Tested on the SeaDronesSee dataset [1] with YOLOv5 [2] and Slicing Aided Hyper Inference (SAHI) framework [3], our approach improves Mean Average Precision (mAP) for small objects by 38% compared to a baseline and achieves more than double the inference speed compared to static tiling. This approach enables more efficient and accurate UAV-based SAR operations under diverse conditions.

Maritime Small Object Detection from UAVs using Deep Learning with Altitude-Aware Dynamic Tiling

TL;DR

Maritime small-object detection from UAV imagery is challenged by low object-to-background pixel ratios at high altitudes. The authors propose altitude-aware dynamic tiling that combines altitude-dependent scaling with adaptive tiling inside the SAHI framework, paired with YOLOv5s, to maintain detection performance while reducing computation. On the SeaDronesSee dataset, the method yields a 38% relative improvement in over a no-tilting baseline and more than a threefold increase in inference speed compared with static tiling, illustrating a favorable accuracy–speed trade-off for real-time SAR operations. The approach offers a practical, altitude-responsive tiling strategy that complements existing detectors and can be extended to multi-modal data or alternative architectures for robust maritime UAV surveillance.

Abstract

Unmanned Aerial Vehicles (UAVs) are crucial in Search and Rescue (SAR) missions due to their ability to monitor vast maritime areas. However, small objects often remain difficult to detect from high altitudes due to low object-to-background pixel ratios. We propose an altitude-aware dynamic tiling method that scales and adaptively subdivides the image into tiles for enhanced small object detection. By integrating altitude-dependent scaling with an adaptive tiling factor, we reduce unnecessary computation while maintaining detection performance. Tested on the SeaDronesSee dataset [1] with YOLOv5 [2] and Slicing Aided Hyper Inference (SAHI) framework [3], our approach improves Mean Average Precision (mAP) for small objects by 38% compared to a baseline and achieves more than double the inference speed compared to static tiling. This approach enables more efficient and accurate UAV-based SAR operations under diverse conditions.

Paper Structure

This paper contains 18 sections, 3 equations, 9 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Proposed Methodology
  4. Adaptive Altitude-Aware Scaling
  5. Tiling
  6. How SAHI works
  7. Proposed Dynamic Tiling for Multiscale Detection
  8. Experimental Setup
  9. Dataset
  10. Dataset Preprocessing
  11. Training sets Preprocessing.
  12. Test set Preprocessing
  13. Model Selection
  14. Training
  15. Results and Discussion
  16. ...and 3 more sections

Figures (9)

  • Figure 1: Illustration of a general tiling process (without overlap), adapted from Raster_Tiling.
  • Figure 2: Dynamic tiling workflow: altitude-aware scaling (DSS) followed by tiling to facilitate object detection.
  • Figure 3: DSS integrated with the SAHI framework and YOLOv5.
  • Figure 4: Dataset Image Altitudes.
  • Figure 5: Dataset Insights (generated by roboflow).
  • ...and 4 more figures