Position Detection and Direction Prediction for Arbitrary-Oriented Ships via Multiscale Rotation Region Convolutional Neural Network

Xue Yang, Hao Sun, Xian Sun, Menglong Yan, Zhi Guo, Kun Fu

Abstract

Ship detection is of great importance and full of challenges in the field of remote sensing. The complexity of application scenarios, the redundancy of detection regions, and the difficulty of dense ship detection are all the main obstacles that limit the successful operation of traditional methods in ship detection. In this paper, we propose a brand new detection model based on a multitask rotational region convolutional neural network to solve the problems above. This model consists of five consecutive parts: Dense Feature Pyramid Network (DFPN), adaptive region of interest (ROI) Align, rotational bounding box regression, prow direction prediction, and rotational non-maximum suppression (R-NMS). First, the low-level location information and high-level semantic information are fully utilized through multiscale feature networks. Then, we design Adaptive ROI Align to obtain high-quality proposals that retain complete spatial and semantic information. Unlike most previous approaches, the prediction obtained by our method is the minimum bounding rectangle of the object with less redundant regions. Therefore, the rotational region detection framework is more suitable to detect dense objects than traditional detection models. Additionally, we can determine the berthing and sailing direction of ships through prediction. A detailed evaluation based on SRSS for rotation detection shows that our detection method has a competitive performance.

Introduction

Detecting ships in remote sensing (satellite) imagery is notoriously difficult due to two primary factors. First, ships possess large aspect ratios; when oriented diagonally, standard horizontal bounding boxes include excessive background noise, which degrades neural network performance. Second, ships are often densely packed in ports. Horizontal boxes for adjacent ships overlap significantly, causing standard Non-Maximum Suppression (NMS) algorithms to erroneously delete valid detections. Furthermore, traditional detectors rarely provide orientation information, which is critical for maritime traffic monitoring.

Proposed Methodology

The proposed Multiscale Rotation Region Convolutional Neural Network addresses these challenges through a specialized architecture designed for arbitrary-oriented object detection.

Architecture of the proposed Multiscale Rotation Region CNN, featuring the Dense Feature Pyramid Network (DFPN) and dual detection branches.

Key Components:

  • Dense Feature Pyramid Network (DFPN): This architecture uses dense connections to fuse low-level positional information with high-level semantic features, ensuring effective detection of both small boats and large tankers.
  • Adaptive ROI Align: While standard ROI Align can include background noise for tilted objects, the authors developed an adaptive version using a trainable mask. This filters out noise while preserving the ship's spatial integrity.
  • Rotational Bounding Box Regression: The model predicts a fifth variable, the angle ($\theta$), in addition to standard coordinates. This allows the bounding box to tilt and tightly encompass the ship's hull.
  • Prow Direction Prediction: A specific branch of the network predicts which side of the bounding box is the prow (front). This enables the system to determine the ship's sailing or berthing orientation.
  • Rotational NMS (R-NMS): A modified suppression algorithm calculates the Intersection-over-Union (IoU) of tilted rectangles, preventing the accidental deletion of closely parked ships.

Experimental Evaluation

The model was evaluated on two primary datasets:

  1. SRSS (Satellite Remote Sensing Ship): A dataset collected by the authors using Google Earth, containing 50 large scenes at 0.5m resolution.
  2. DOTA: A large-scale public dataset for arbitrary-oriented object detection in aerial images, serving as the gold standard for benchmarking.

Results and Analysis

The rotational detection framework demonstrated significant improvements over traditional horizontal methods, particularly in dense port environments.

Detection results using the rotational region framework, showing tight bounding boxes and prow direction indicators.

Performance Metrics:

  • DOTA Performance: The model achieved a Mean Average Precision (mAP) of 62.29, significantly outperforming SSD (17.84), YOLOv2 (25.49), and standard Faster R-CNN (39.95).
  • Ship-Specific Accuracy: On the SRSS dataset, the model reached an F1-score of 84.9%.
  • Directional Accuracy: Prow prediction accuracy closely tracked the overall recall rate, indicating that the model reliably identifies orientation whenever a ship is successfully detected.

Conclusion

By utilizing rotational bounding boxes and a dedicated prow prediction branch, this model effectively handles the challenges of large aspect ratios and dense packing in maritime remote sensing. While the model is highly accurate, future work will focus on reducing false alarms caused by complex coastal structures and optimizing processing speed for real-time applications.

Summary updated

Abstract

Ship detection is of great importance and full of challenges in the field of remote sensing. The complexity of application scenarios, the redundancy of detection region, and the difficulty of dense ship detection are all the main obstacles that limit the successful operation of traditional methods in ship detection. In this paper, we propose a brand new detection model based on multitask rotational region convolutional neural network to solve the problems above. This model is mainly consist of five consecutive parts: Dense Feature Pyramid Network (DFPN), adaptive region of interest (ROI) Align, rotational bounding box regression, prow direction prediction and rotational nonmaximum suppression (R-NMS). First of all, the low-level location information and high-level semantic information are fully utilized through multiscale feature networks. Then, we design Adaptive ROI Align to obtain high quality proposals which remain complete spatial and semantic information. Unlike most previous approaches, the prediction obtained by our method is the minimum bounding rectangle of the object with less redundant regions. Therefore, rotational region detection framework is more suitable to detect the dense object than traditional detection model. Additionally, we can find the berthing and sailing direction of ship through prediction. A detailed evaluation based on SRSS for rotation detection shows that our detection method has a competitive performance.
BibTeX
@inproceedings{yang2018position,
  author = {Xue Yang and Hao Sun and Xian Sun and Menglong Yan and Zhi Guo and Kun Fu},
  title = {{Position Detection and Direction Prediction for Arbitrary-Oriented Ships via Multiscale Rotation Region Convolutional Neural Network}},
  booktitle = {IEEE Access},
  year = {2018},
  url = {https://openreview.net/forum?id=-n__F0oxOMn},
}
Figure