Enhanced Feature Fusion for Small Ship Segmentation
Author Information
Author(s): Rabi Sharma, Muhammad Saqib, Lin C. T., Michael Blumenstein
Primary Institution: School of Computer Science, University of Technology Sydney
Hypothesis
The study hypothesizes that enhancing different layers of semantic information during feature extraction will improve the segmentation of small ships in complex marine environments.
Conclusion
The proposed enhanced ASPP feature fusion module significantly improves the accuracy of small ship instance segmentation compared to existing methods.
Supporting Evidence
- The enhanced ASPP feature fusion module outperformed state-of-the-art models in three diverse datasets.
- Average precision scores for small ship segmentation improved significantly with the proposed method.
- The method effectively captures fine details of small ships in complex backgrounds.
Takeaway
This study helps computers better see and identify small ships in the ocean, even when they are far away or hard to see.
Methodology
The study uses a novel enhanced ASPP feature fusion module integrated into a Cascade Mask R-CNN framework, tested on three maritime datasets.
Limitations
The method increases computational demands, which may affect real-time performance.
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website