A model simplification method is proposed to accelerate two-stage object detectors for on-device inference in remote sensing, by utilizing only one feature extraction and applying a high-pass filter, while maintaining accuracy.
This survey provides a comprehensive overview of recent advancements in oriented object detection in optical remote sensing images using deep learning techniques. It highlights the key challenges, including feature misalignment, spatial misalignment, and periodicity of angle, and discusses how existing methods address these issues through detection frameworks, oriented bounding box regression, and feature representations.
The proposed Location Refined Feature Pyramid Network (LR-FPN) enhances the extraction of shallow positional information and facilitates fine-grained context interaction to improve remote sensing object detection performance.