Wide-range and high-resolution imaging is one of the important development directions of remote sensing cameras. In order to achieve wide-range and high-resolution imaging, the most common method is to scan a large field of view through a scanning mirror. As an important component of remote sensing cameras to achieve large field of view imaging, scanning mirrors are usually located at the forefront of the entire imaging system, and expand the imaging field of view by changing the incident optical angle. It is easily to expose the scanning mirror to the direct influence of external heat flow, such as sunlight and Earth's reflected light, which may lead to change in the temperature of the scanning mirror, and the thermal stress can lead to change in the surface figure of the scanning mirror, thereby affecting the imaging quality of the camera and limiting the available time of the camera in orbit. Therefore, when more resources cannot be provided to ensure the temperature stability of the scanning mirror, the best way to achieve high-quality imaging is to compensate for surface change caused by thermal deformation. How to accurately test the surface change of the scanning mirror in orbit becomes a key step in whether to compensate for thermal deformation surface change.The proposed method is based on an algorithm that combines the interferometer data and the temperature of the mirror to establish a relationship model between the temperature of the mirror and the deformation of the mirror surface. In this paper, a neural network algorithm is adopted, which puts the temperature measurement values of the mirror as the input and the surface figure change of the mirror as the output (Fig.4). The function of inverting the surface mirror change of the mirror through the temperature values of the mirror can be realized. The input temperature values are provided by eight temperature measuring points on the back of the mirror (Fig.7). The surface figure of the mirror is expressed using a Zernike expression, including Power, Astigmatism, Coma, Trefoil, and Spherical aberration (Tab.5).The method can achieve better performance than RMS 12.6 nm (Fig.8), which is a good performance. The core of the algorithm is how to establish the relationship between the temperature of the mirror and the surface figure change. Therefore, this paper proposes to use the neural network algorithm to reverse the surface figure change of the mirror based on the temperature measurement points on the back of the mirror, providing a basis for the use of deformable mirror in the rear optical path for correction. This method eliminates the need for additional wavefront testing systems, which can reduce the implementation difficulty of optical systems, as well as additional load burdens such as weight and power consumption. And The method is also more conducive to in-orbit applications.The scanning mirror of a remote sensing camera has a significant impact on the imaging quality under the influence of external heat flow. In order to achieve more efficient imaging compensation in orbit, this paper proposes a method of using neural network algorithms to invert the surface figure change of the scanning mirror through the temperature measurement points of the scanning mirror itself. The method can accurately reflect the surface figure change of the scanning mirror caused by external heat flow, thereby providing accurate surface figure change for the compensation of the optical path after implementation. Through analysis, The residual errors between the retrieved surface figure and the theoretical surface figure is less than 12.6 nm, providing a new solution for remote sensing cameras to achieve better surface figure test method in orbit and to compensate for the impact of external heat flow.