In ground-based optical detection systems for mid-high orbit targets during daytime， the extremely low SNR （Signal-to-Noise Ratio） of mid-high orbit imaging necessitates the use of multi-frame joint correlation detection techniques for target extraction. To address the challenge of target extraction caused by stellar paths， this paper introduces a stars removal algorithm for single-frame images. The process involves generating a background image with replaceable stars from the original image， identifying suspected targets in a single frame based on stellar imaging characteristics， detecting real stellar targets through single-frame and multi-frame association methods， and replacing stellar targets in the original image with corresponding background pixels. Experimental results show the algorithm effectively removes stars with apparent magnitudes of 4 to 9， achieving an average removal success rate of 98%. After multi-frame joint correlation detection， target intensity remains unchanged， while stellar paths are significantly reduced. The algorithm processes a full-frame image in under 12ms， meeting the real-time requirements of a 25 Hz camera frame rate. This approach significantly enhances the extraction of mid-high orbit targets using multi-frame joint correlation detection.