Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Sparse R-CNN OBB: Ship Target Detection in SAR Images Based on Oriented Sparse Proposals

Kamirul Kamirul, Odysseas Pappas, Alin Achim

TL;DR

The paper tackles oriented ship detection in SAR imagery, a challenging task due to arbitrary ship orientations and clutter. It introduces Sparse R-CNN OBB, a rotated extension of Sparse R-CNN that uses 300 sparse learnable proposals and Rotated RoIAlign, coupled with a six-head refinement mechanism, to detect oriented ships without dense anchors. Across the RSDD-SAR dataset, it achieves state-of-the-art AP50 scores, with offshore performance exceeding 96% and overall gains over most baselines, while maintaining a simpler, end-to-end trainable architecture. The work demonstrates that sparse proposals can deliver high accuracy and efficiency for SAR ship detection, enabling robust surveillance in maritime environments.

Abstract

We present Sparse R-CNN OBB, a novel framework for the detection of oriented objects in SAR images leveraging sparse learnable proposals. The Sparse R-CNN OBB has streamlined architecture and ease of training as it utilizes a sparse set of 300 proposals instead of training a proposals generator on hundreds of thousands of anchors. To the best of our knowledge, Sparse R-CNN OBB is the first to adopt the concept of sparse learnable proposals for the detection of oriented objects, as well as for the detection of ships in Synthetic Aperture Radar (SAR) images. The detection head of the baseline model, Sparse R-CNN, is re-designed to enable the model to capture object orientation. We train the model on RSDD-SAR dataset and provide a performance comparison to state-of-the-art models. Experimental results show that Sparse R-CNN OBB achieves outstanding performance, surpassing most models on both inshore and offshore scenarios. The code is available at: www.github.com/ka-mirul/Sparse-R-CNN-OBB.

Sparse R-CNN OBB: Ship Target Detection in SAR Images Based on Oriented Sparse Proposals

TL;DR

The paper tackles oriented ship detection in SAR imagery, a challenging task due to arbitrary ship orientations and clutter. It introduces Sparse R-CNN OBB, a rotated extension of Sparse R-CNN that uses 300 sparse learnable proposals and Rotated RoIAlign, coupled with a six-head refinement mechanism, to detect oriented ships without dense anchors. Across the RSDD-SAR dataset, it achieves state-of-the-art AP50 scores, with offshore performance exceeding 96% and overall gains over most baselines, while maintaining a simpler, end-to-end trainable architecture. The work demonstrates that sparse proposals can deliver high accuracy and efficiency for SAR ship detection, enabling robust surveillance in maritime environments.

Abstract

We present Sparse R-CNN OBB, a novel framework for the detection of oriented objects in SAR images leveraging sparse learnable proposals. The Sparse R-CNN OBB has streamlined architecture and ease of training as it utilizes a sparse set of 300 proposals instead of training a proposals generator on hundreds of thousands of anchors. To the best of our knowledge, Sparse R-CNN OBB is the first to adopt the concept of sparse learnable proposals for the detection of oriented objects, as well as for the detection of ships in Synthetic Aperture Radar (SAR) images. The detection head of the baseline model, Sparse R-CNN, is re-designed to enable the model to capture object orientation. We train the model on RSDD-SAR dataset and provide a performance comparison to state-of-the-art models. Experimental results show that Sparse R-CNN OBB achieves outstanding performance, surpassing most models on both inshore and offshore scenarios. The code is available at: www.github.com/ka-mirul/Sparse-R-CNN-OBB.
Paper Structure (16 sections, 5 equations, 3 figures, 2 tables)

This paper contains 16 sections, 5 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Sparse Learnable Proposals
  4. Dynamic Head
  5. Sparse R-CNN OBB
  6. Backbone
  7. Rotated Sparse Learnable Proposals
  8. Loss Function
  9. Experiment Setup
  10. Dataset
  11. Evaluation
  12. Hyperparameters and Environment
  13. Results and Discussion
  14. Effect of Number of Proposals
  15. Comparison to SOTA Models
  16. ...and 1 more sections

Figures (3)

  • Figure 1: Oriented bounding box representation used in the proposed Sparse R-CNN OBB.
  • Figure 2: Flowchart of the proposed Sparse R-CNN OBB.
  • Figure 3: Detection results across different environments, with predictions (magenta) and ground truth (green). The first two left columns show results in inshore regions, while the remaining columns correspond to offshore regions.