Abstract: Highlighting the circular structure of the cone-drogue with optical markers is widely used for safe autonomous aerial refueling (AAR) docking. However, the complex interference scenarios such as strong sunlight, high-radiance clouds, and cone-drogue occlusions significantly reduce the accuracy of cone-drogue recognition and pose estimation, and may even prevent reliable localization. To address these complex interference scenarios, a cone-drogue recognition and pose estimation method based on cone-drogue imaging characteristics is proposed. Firstly, the morphological filtering with circular structuring elements is applied to suppress irregular strong sunlight and high-radiance clouds based on the circular imaging features of the LED markers, thereby enhancing the signal-to-noise ratio and the extraction accuracy of the LED markers. Subsequently, the cone-drogue is successfully recognized from a large amount of background interference by combining spatial characteristics such as the circularity of LED markers, inter-marker distances and marker-count constraints, as well as the consistency of the cone-drogue marker cluster (e.g., area similarity and angular uniformity). Finally, a pose estimation algorithm based on perspective-n-point (PnP) optimization with reprojection error minimization is proposed to accurately estimate the cone-drogue pose and enhance its occlusion robustness. Ground-based experimental results on the cone-drogue measurements demonstrate that the proposed method significantly improves the accuracy and anti-interference capability of cone-drogue recognition and pose estimation under complex interference backgrounds such as strong sunlight, high-radiance dense clouds, and cone-drogue occlusion compared to state-of-the-art algorithms. Specifically, the measurement errors of distance and azimuth are less than 3.5 cm and 0.2 cm, respectively, at a distance of 10 m. The measurement errors of distance and azimuth are less than 0.15 cm and 0.10 cm, respectively, at a distance of 2 m. The recognition accuracy exceeds 95% when the cone-drogue occlusion area reaches 60%, with an average processing time of 16.0 ms. The method could be employed in autonomous aerial refueling docking to increase the success rate of AAR operations by virtue of high accuracy, real-time performance, and scene adaptability.