<p>To enhance the robustness of pose estimation for non-cooperative space targets under complex lighting conditions, this paper proposes a three-stage adaptive pose estimation algorithm that combines deep learning-based object detection with high-precision pose estimation. First, deep learning is employed to identify and extract the target’s contour features while eliminating background interference. Second, the detection accuracy of the target’s contour line segments is effectively enhanced by integrating an improved line segment detector, resulting in feature points with higher precision. Finally, a pose estimation algorithm which combines two perspective-n-point methods is proposed for these feature points, and the estimation results are optimized using a Kalman filter. Simulation experiment results demonstrate that the proposed algorithm can adaptively estimate the pose of different targets, effectively improving pose estimation accuracy under complex lighting conditions, while reducing computation time.</p>