Automatic Ship Detection of Remote Sensing Images from Google Earth in Complex Scenes Based on Multi-Scale Rotation Dense Feature Pyramid Networks (1806.04331v1)

Abstract: Ship detection has been playing a significant role in the field of remote sensing for a long time but it is still full of challenges. The main limitations of traditional ship detection methods usually lie in the complexity of application scenarios, the difficulty of intensive object detection and the redundancy of detection region. In order to solve such problems above, we propose a framework called Rotation Dense Feature Pyramid Networks (R-DFPN) which can effectively detect ship in different scenes including ocean and port. Specifically, we put forward the Dense Feature Pyramid Network (DFPN), which is aimed at solving the problem resulted from the narrow width of the ship. Compared with previous multi-scale detectors such as Feature Pyramid Network (FPN), DFPN builds the high-level semantic feature-maps for all scales by means of dense connections, through which enhances the feature propagation and encourages the feature reuse. Additionally, in the case of ship rotation and dense arrangement, we design a rotation anchor strategy to predict the minimum circumscribed rectangle of the object so as to reduce the redundant detection region and improve the recall. Furthermore, we also propose multi-scale ROI Align for the purpose of maintaining the completeness of semantic and spatial information. Experiments based on remote sensing images from Google Earth for ship detection show that our detection method based on R-DFPN representation has a state-of-the-art performance.

• The paper presents a novel R-DFPN framework that significantly improves ship detection in complex remote sensing scenes.
• It leverages dense feature pyramid networks, rotation anchors, and multi-scale ROI Align to handle varied ship scales and poses accurately.
• Experimental results show high performance with 88.2% recall, 91.0% precision, and 89.6% F-measure, outperforming traditional methods.

Overview of Automatic Ship Detection in Complex Remote Sensing Images

The paper, "Automatic Ship Detection of Remote Sensing Images from Google Earth in Complex Scenes Based on Multi-Scale Rotation Dense Feature Pyramid Networks," introduces a novel framework termed Rotation Dense Feature Pyramid Networks (R-DFPN). This framework targets the complex challenge of ship detection in remote sensing images from Google Earth, addressing limitations inherent in traditional detection methods and improving object detection performance in complex maritime environments.

Framework and Methodology

R-DFPN is designed with several key components to enhance detection capabilities:

1. Dense Feature Pyramid Networks (DFPN): Unlike traditional Feature Pyramid Networks (FPN) that may fall short in feature propagation and reuse, DFPN leverages dense connections to generate high-level semantic feature maps across all scales. This enhancement supports the detection of ships with varied scales and aspect ratios.
2. Rotation Anchors: A critical innovation in R-DFPN is the use of rotation anchors to predict ship poses more accurately. This approach mitigates the problematic redundancy and overlap in detection regions typical of horizontally anchored methods.
3. Multi-Scale ROI Align: To maintain precise semantic and spatial information, the framework introduces multi-scale ROI Align. This technique addresses the feature misalignment issues typical in ROI pooling, crucial for handling ships with high aspect ratios.

Experimental Evaluation

The experiments conducted demonstrate that R-DFPN achieves state-of-the-art performance in ship detection tasks, demonstrating significant results: 88.2% Recall, 91.0% Precision, and 89.6% F-measure. The research compares various detection methods, highlighting the superiority of R-DFPN. For instance, Faster-RCNN and traditional FPN suffer from lower Recall due to non-optimal handling of dense ship arrangements and misalignment issues.

Key Contributions

The primary contributions of this research paper are:

• Developing a rotating region-based detection framework to handle complex and densely populated maritime scenes.
• Enhancing feature propagation and reuse through the dense connections in DFPN.
• Introducing rotation anchors to reduce non-maximum suppression effects and improve detection recall.
• Implementing multi-scale ROI Align for improved semantic and spatial information retention, addressing typical ROI pooling issues.

Implications and Future Developments

Practically, this method can enhance national defense, port management, and maritime rescue operations by providing reliable ship detection in challenging environments. Theoretically, it contributes to enhancing multi-scale object detection architectures, particularly in complex scene analysis in AI.

Looking forward, the research could focus on reducing false alarms caused by environmental interference, such as roofs or docks, that mimic ship-like features. Exploring enhanced sea-land segmentation techniques or leveraging generative adversarial networks might bolster R-DFPN's robustness against such confusions.

In conclusion, the R-DFPN framework presents a significant advancement in tackling the nuanced challenges of ship detection in complex remote sensing scenes, suggesting promising avenues for future research and applications.