With the continuous development of the field of remote sensing, space-based remote sensing is developing in the direction of all-sky and intelligent. Since low-light remote sensing is used to detect ground objects under low illumination conditions such as night and morning and night periods, it results in the characteristics of low contrast, low brightness and low signal-to-noise ratio of remote sensing images, among which, low signal-to-noise ratio leads to a large number of complex physical noises drowning the image features, seriously affecting the recognition and interpretation of ground objects. This paper summarizes the actual full-link physical model based on optical remote sensing imaging and the technical approaches to improve the signal-to-noise ratio of remote sensing images, and summarizes the methods based on traditional filtering, physical model and deep learning respectively. By comparing the differences among the main representative algorithms of various methods, the paper summarizes their respective characteristics. The future development direction of the improvement of the signal-to-noise ratio of space-based remote sensing images is forecasted.