Anchor-free remote sensing object detection method based on multi-scale features and angle information

Remote sensing object detection plays a crucial role in interpreting remote sensing imagery, yet it faces challenges such as significant scale variations and diverse object orientations. To address these issues, this paper proposes an anchor-free detection method that incorporates multi-scale features and angular information. A feature selection and alignment module is integrated into a feature pyramid network to enhance multi-scale representation learning and alleviate feature misalignment. Furthermore, a rotation bounding box localization method is introduced within the anchor-free framework, eliminating anchor-related hyper-parameters and improving detection robustness. To mitigate boundary discontinuities in rotated detection, bounding boxes are modeled as two-dimensional Gaussian distributions, and a novel rotation-sensitive regression loss is proposed. Experimental results demonstrate that the proposed method outperforms state-of-the-art approaches, particularly in detecting multi-scale and rotated objects.