With the rapid development of optical remote sensing satellites in China, the types of satellites, imaging methods, and the number of satellites in orbit are constantly increasing. Intelligent remote sensing satellite (IRSS) and related surveying and mapping application services have become an important development direction in the field of remote sensing surveying and mapping. As a key technology, on-orbit geometric positioning is rapidly developed and has become one of the essential works for IRSS and its subsequent applications. To improve the positioning accuracy, at the same time, ground control information has to be injected into the spaceborne processing system. However, the data volume of ground control information is usually globally covered and is too huge to be directly loaded. How to lighten the ground control information, therefore, has become one of the urgent problems to be solved. The construction of an on-orbit control information library is one of the solutions. At present, most related methods are focused on the lightweight processing of control information descriptors, the performance of which is generally limited by environmental noise-affected feature diversity. To address the problem, scholars have proposed and attempted to introduce vector line features into the methods to improve its robustness. Based on this, we propose an optical remote sensing satellite positioning technology based on the lightweight vector control library, which uses the vector topology relationship to get rid of the dependence of the control point on local image grayscale information. We believe that our spaceborne lightweight vector control library can help IRSS improve on-orbit positioning accuracy and advance the development of on-orbit automatic processing technology.

First, we extract and perfect the road information on the ground platform. Through a series of operations including road network refinement, node interruption, and road network topological relationship construction, the spaceborne lightweight vector control library is formed. Then, the road mask is extracted from the spaceborne intelligent processing module. After that, in the matching and positioning module, we propose the local vector construction method based on random walking for the unavoidable breakage phenomenon of road extraction in the spaceborne environment. Finally, we propose a hierarchical matching strategy based on the hidden Markov model (HMM) to accomplish the matching and use the offset of the matching transformation model to correct its rational polynomial coefficients (RPC) to improve the image positioning accuracy.

To verify the effectiveness of the proposed method, we select three sets of data for testing: 1) high score No. 2 image and high score No. 7 image in Kaifeng City, Henan Province, China; 2) Jilin-1 video data and Google images in Juarez, USA; 3) Jilin-1 video data and Google images in Oklahoma City, USA. Random walk vector extraction is performed multiple times in the three sets of data, and the average number of vector nodes extracted is 13, 7, and 5 respectively (Table 1). In addition, we compare the storage capacity of this method with traditional image control points, lightweight image control points, and deep lightweight image control points method. After comparison, the storage capacity of this method is reduced by 83334.0 times, 252.6 times, and 2.3 times compared to that of the other three methods (Table 2). In the vector matching stage, the proposed methods successfully match the lightweight vector library and the random walk vector. The number of vector nodes successfully matched in the three groups of regions is 17, 21, and 5, and the encryption points are 4372, 4245, and 5591 respectively. The test results all meet the minimum requirements for the number of points for the control point to correct the image positioning accuracy. Therefore, our method is feasible and effective for correcting the image positioning accuracy. Then, we use the successfully matched vector nodes and encryption points to correct the image RPC and compare the positioning accuracy of the original image. The results show that by correcting the RPC through the proposed method, the image localization accuracy of each region is significantly improved. Specifically, the positioning accuracy of Jilin No. 2 has been increased from 79 pixel to about 4 pixel, that of Jilin-1 video data in region 2 has been increased from 675 pixel to about 16 pixel, and that of Jilin-1 video data in region 3 has been increased from 539 pixel to about 17 pixel (Table 2). Through above tests, it is verified that the proposed method can effectively improve the positioning accuracy of non-measurable optical remote sensing satellite images.

Given the urgent demand for spaceborne lightweight control data required of intelligent remote sensing satellite systems, we design a spaceborne lightweight vector control library. Compared with traditional control point methods, the proposed vector control library uses only the geographic coordinates and topological relationship of the control points to complete the image positioning without descriptors of local images or image points. Experiments such as vector construction, vector matching, and positioning accuracy comparison are conducted on the proposed method through multiple sets of image data to verify the matching effectiveness of the lightweight vector control library and its ability to improve positioning performance. This leads to the conclusion that the lightweight vector control library can provide high-precision global positioning information for intelligent remote sensing satellite systems and provide data support for applications such as on-orbit earth positioning and geometric correction.