The automation of calibration and the accuracy of pixel extraction are the key factors in improving the efficiency and accuracy of geometric calibration of optical remote sensing satellites in orbit. With the increasing number of optical remote sensing satellites in orbit and the continuous improvement of satellite image resolution, current level of calibration automation cannot meet future development needs. The coordinates of control points and their corresponding image points are input parameters for the geometric calibration of the camera. The accuracy of their measurement and extraction directly affects the accuracy and precision of the internal and external orientation element calibration of the satellite camera. The measurement accuracy of ground control points and on-orbit attitude has reached centimeter or even millimeter-level precision. However, the pixel extraction of control point images mainly relies on manual selection or pixel measurement algorithms, which cannot match the accuracy of control point measurements. Lightweight and automated quasi-stellar source-reflected point sources are proposed to be used as ground control points, along with an algorithm for matching pixel extraction from these sources. This will be combined with on-orbit calibration experiments based on reflected point sources to verify the accuracy of pixel extraction for optical remote sensing satellites. The objective of this paper is to improve the accuracy of pixel extraction, which in turn will enhance the geometric calibration and remote sensing image positioning accuracy of these satellites in orbit.

First, we propose a Gaussian model-based method for pixel extraction from reflected point sources and a rational function model-based method for verifying the accuracy of pixel extraction. Based on these principles, we design an experiment to verify the accuracy of pixel extraction from reflected point sources. Second, we analyse the experimental image and extract the pixels by the reflected point source method. The accuracy of the extracted pixels is then verified by template matching and nonlinear optimization methods. Finally, we verify the accuracy of pixel extraction by the reflected point source method with a rational function model and calculate the accuracy of pixel positioning by the template matching method and nonlinear optimization method. A comparative verification of the reflected point source method is conducted.

The pixel coordinates extracted by the reflected point source method, template matching method, and nonlinear optimization method are shown in Tables 1 and 4. The results of the three methods are relatively consistent. The differences between the extracted pixel coordinates and the mean values are shown in Table 5. The maximum error is 0.0472 pixel, which is better than 0.05 pixel, which verify the accuracy of the reflected point source method for the optical remote sensing satellite. The pixel extraction accuracy is better than that of square-shaped top-hat markers and manual extraction of natural features. The validation results of the rational function model are shown in Table 7. After the maximum value is removed, the positioning accuracy of the extracted pixel coordinates using the reflected point source method, template matching method, and nonlinear optimization method in the flight direction and sensor line array direction is better than 0.04 pixel, and the planar accuracy is better than 0.05 pixel, further verify the accuracy of pixel extraction using the reflected point source method. The experimental results preliminarily indicate that the reflected point source method can achieve high-precision remote sensing image pixel extraction and meet the application requirements of high-precision targets for the geometric calibration of optical remote sensing satellites.

The preliminary results of on-orbit tests based on reflected point sources for optical remote sensing satellites indicate that compared to conventional methods such as template matching, the reflected point source method has an extraction accuracy of better than 0.05 pixel, which is higher than that for natural objects and artificial targets. The rational function model parameters accompanying the remote sensing images are applied to verify the positioning accuracy of reflected point source method, which is better than 0.05 pixel. Automated reflected point source method not only enable high-precision pixel extraction but also combine the array configuration of point sources with the automatic measurement of atmospheric optical characteristics in the test area. Image quality evaluation and radiation calibration are achieved for optical remote sensing satellites. This can improve the efficiency and accuracy of on-orbit calibration for optical remote sensing satellites. Automated reflected point sources are expected to be networked and deployed within the field of view of optical remote sensing satellites, forming ground control points for on-orbit geometric calibration and enabling high-precision, high-frequency, and normalized geometric radiation calibration for optical remote sensing satellites. Our study provides a reference for automated calibration and accuracy improvement of domestic remote sensing satellites and is of great significance for improving the quantitative level and application of optical remote sensing satellite images.