Low-light satellite can capture the information of town lights, fishing boat lights, and fire points to form nighttime light remote sensing images (hereafter referred to as nighttime light image, NTL image). High-resolution NTL images help to reflect human activities on the ground more clearly. It has been widely used in urban expansion, economic evaluation, carbon emission analysis, and other fields, and has become an important data source for natural and social science research. The change analysis based on multi-temporal NTL image is of great significance for monitoring human ground activities and urban development, and geometric correction of multi-temporal remote sensing image can guarantee the spatial alignment of the multi-temporal image, which is the basis for change analysis of multi-temporal NTL image. Due to the characteristics of night imaging of NTL images, there is lack of geometric reference for NTL images at home and abroad, so few scholars have carried out geometric correction research on NTL images. According to the imaging characteristics of NTL images, this paper proposes to take high-precision road network data as geometric reference, extract control points by automatic matching between NTL images and road network data, and then realize geometric correction of NTL images. Taking the Luojia 1-01 (LJ 1-01) satellite NTL image as an example, the experimental verification shows that the accuracy of geometric correction based on road network control can reach the sub-pixel level, which verifies the feasibility of the proposed method. This paper verifies the rationality of using the road network as the benchmark data for NTL images, provides a feasible idea for subsequent scholars to study the geometric processing of NTL images, and ensures the geometric quality of data for the application of multi-temporal NTL images.