| This paper discusses the problems and improvement methods of the algorithms of star sensors in the process of star identification under the condition of dynamic change of attitude angle for the spacecraft.Attitude determination is the premise of stable and precise control of spacecraft.Attitude measurement under high dynamic conditions re-quires faster speed and higher accuracy.Therefore,the development and application of high dynamic star sensors has become the focus of our field.Under high dynamics,the energy of the stars mapped on the sensor is dispersed into different pixels,so that the star points are drowned in the background and noise,or form a smearing shape on the image.These lead to the decrease of the positioning accuracy of the star point extrac-tion algorithm of the star sensor and the decrease of the rate of the star identification algorithm.Aiming at the problems,this paper studies the formation and principle of star images under dynamic conditions,the algorithm of star point extraction and the algorithm of star identification.The focus is from the perspective of image processing and image recognition,From the perspective of image processing and image recogni-tion,the characteristics of algorithms are analyzed and the improvement research of the algorithms is carried out,in order to improve the accuracy of the star sensor under high dynamic conditions.Therefore,the research content of this paper mainly includes the following three aspects:（1）The principle and model of star imaging under dynamic conditions of star sen-sor are studied,and the causes and influencing factors of smearing phenomenon are analyzed from the perspective of image.First of all,by studying the mapping process of stars on the star sensor detector,this paper accurately establishes the attitude map-ping model and energy model of static imaging.Secondly,combined with the dynamic attitude motion parameters of the star sensor,the above two models are modified un-der the dynamic condition respectively,so that the dynamic star point imaging model is given.Finally,through the dynamic star image simulation,the characteristics of trailing in different motion states are discussed,and the prior knowledge on the image is given d on the motion prior information.In this way,the smearing star image under dynamic conditions is obtained,which provides a basis for the verification of the subsequent algorithm performance.（2）For the smearing star images generated by the star sensor under dynamic con-ditions,this paper studies various image processing algorithms.And aiming at the low accuracy and poor precision of the extraction algorithm,a star point extraction algo-rithm d on image matching is proposed.First,this paper studies the preprocessing for star images,the recovery of star point energy and the extraction of star point informa-tion.As the signal-to-noise ratio of star points is seriously reduced and the smearing phenomenon occurs,the denoising algorithm is not effective,the recovery speed of star points is slow and easily disturbed by noise,and the accuracy of star point extraction and positioning is reduced.In order to solve the above problems,this paper proposes a star point extraction method d on image matching with the prior knowledge of dynamic star images.d on the principle of cross-correlation,the dynamically degraded star im-age is matched with the fuzzy kernel,which can achieve accurate star point extraction under noise and non-uniform background interference.Finally,through simulation ex-periments,the improved algorithm has a discovery rate of 58%,and it is verified that the algorithm improves the accuracy and running speed.（3）For the identification algorithm of star images under dynamic conditions,this paper proposes an end-to-end algorithm for direct identification of smearing star image.Due to the poor positioning accuracy of the star points in the dynamic star images,the accuracy of the identification algorithm is seriously reduced.Therefore,based on the attention mechanism of neural network,this paper encodes the position information at star points in the images,and focuses on extracting relative position features.And combined with the strong robustness of network classification,a recognition network for smearing star images is proposed.The star points extraction is integrated into the network,and the smearing star image is directly identified.Finally,through simulation experiments,the strong robustness of the network to position noise is verified.It has strong robustness within 10°/S.When the position error is 5 pixels,the recognition rate can still reach 90%,and the identification rate of smearing star images is improved under different attitude angular velocities. |
PDF Full Text Request