The optical detection products for space targets are an important part of satellite engineering， and their reliability and accuracy are crucial factors in determining the effectiveness of on-orbit missions. However， stray light can seriously affect the on-orbit performance of such products. Moreover， the existing anti-stray light algorithms usually have limited feature extraction and high false alarm rates， or complex processing and large computational load， and are not suitable for real-time on-orbit processing. In this paper， a stray light filtering algorithm for space object detection is proposed. The algorithm extracts three types of feature quantities based on independent image blocks， including grayscale mean， intra-block grayscale gradient， and maximum intra-block grayscale difference between classes. It determines the degree of stray light interference， takes the corresponding filtering measures， and finally filters out the effective image data containing the target. In addition， this algorithm can achieve parallel computing along with image data streams through hardware， with a simple calculation process and excellent real-time performance. It is convenient to be embedded in products as embedded modules. The results of simulation and outdoor star observation experiments indicate that the algorithm has exceedingly good performance in filtering stray light. It has stable and reliable target detection capabilities even under stray light interference， and has a low false alarm rate. The algorithm can significantly improve the anti-stray light ability and on-orbit reliability of space target detection products such as star sensors and wide-field space target surveillance cameras.