Rotation-Sensitive Feature Enhancement Network for Oriented Object Detection in Remote Sensing Images.

Area of Science:

• Computer Vision
• Remote Sensing Image Analysis
• Machine Learning

Background:

• Oriented object detection in remote sensing is challenging due to arbitrary rotations, scale variations, and complex backgrounds.
• Existing rotated detectors suffer from insufficient orientation-sensitive features, feature misalignment, and unstable rotation parameter optimization.

Purpose of the Study:

• To propose an enhanced Rotation-Sensitive Feature Pyramid Network (RSFPN) to overcome limitations in current rotated object detectors.
• To improve the accuracy and efficiency of oriented object detection in remote sensing imagery.

Main Methods:

• Introduced a Dynamic Adaptive Feature Pyramid Network (DAFPN) for bidirectional multi-scale feature fusion.
• Developed an Angle-Aware Collaborative Attention (AACA) module using orientation priors for feature refinement.
• Implemented a Geometrically Consistent Multi-Task Loss (GC-MTL) for unified rotation parameter regression with smoothing and adaptive weighting.

Main Results:

• Achieved state-of-the-art mean Average Precision (mAP) of 77.42% on DOTA-v1.0 and 91.85% on HRSC2016.
• Maintained efficient inference speed at 14.5 FPS, demonstrating a strong accuracy-efficiency trade-off.
• Visual analysis confirmed concentrated, rotation-aware feature responses and effective background suppression.

Conclusions:

• The proposed RSFPN framework offers a robust solution for detecting multi-oriented objects in high-resolution remote sensing images.
• The method holds significant practical value for applications such as urban planning, environmental monitoring, and security.