A Method for Determining Satellite Attitude and Rotation Information Based on Deep Learning Target Recognition Algorithm

Abstract

The invention discloses a method for determining satellite attitude rotation information based on a deep learning target recognition algorithm. The method mainly includes: on-orbit acquisition of a satellite three-dimensional model; training of a target recognition algorithm based on deep learning; relative attitude recognition based on deep learning; rotation Information OK. The present invention realizes the determination method of the satellite attitude rotation information based on the deep learning target recognition algorithm, realizes the determination of the current attitude angle of the satellite through the two-dimensional image information of the satellite, and calculates the current satellite motion state through the attitude angle of several frames of images, so that the attitude recognition is performed by the satellite. The recognition problem under the 3D model is transformed into a 2D problem, which simplifies the recognition process, reduces the complexity of the recognition algorithm, and improves the recognition efficiency. Moreover, since this method does not need artificial extraction of satellite features, it is suitable for both cooperative and non-cooperative targets. Due to its versatility, this method has important engineering application value for satellite space target recognition tasks.

Description

technical field [0001] The invention relates to a method for determining satellite attitude rotation information based on a deep learning target recognition algorithm, and belongs to the field of intelligent target recognition of space vehicles. Background technique [0002] At present, the methods for intelligent identification of satellites in orbit mainly include two types. [0003] For the identification of cooperative targets, including the measurement markers (including luminous markers or corner reflectors) specially designed for identification on the spacecraft; the measurement sensors used for relative navigation mainly include microwave rendezvous radar, laser rendezvous radar, relative GPS , visual sensor, laser rangefinder and other equipment, in use, choose a certain sensor or a combination of several sensors. [0004] For the identification of non-cooperative targets, including flash ranging imaging radar without scanning mechanism, the radar can obtain panora...

AI Technical Summary

Problems solved by technology

[0005] For the above-mentioned traditional on-orbit target recognition method, the recognition process is difficult, the cost is high, it is difficult to meet the real-time attitude recognition on-orbit, and the versatility is poor

For the recognition of cooperative targets, there are many mechanical structure designs in the early stage, and special auxiliary measurement markers need to be designed and installed, and the recognition algorithm is also aimed at specific structures; for non-cooperative targets without specific structures, binocular stereo vision cameras and flashlights are required The motion information can only be obtained after three-dimensional spatial modeling of payload means such as ranging radar, and the process is complicated

Images