论文部分内容阅读
为了实现星敏感器对航天器当前姿态的准确测量,如何提高星图识别算法的实时性和鲁棒性成为星敏感器的关键技术。对星图识别过程中应用的模式提取、训练样本集的建立以及神经网络训练方式的改进等算法进行研究。首先,设计一种基于星图特征的三角形剖分方法,将视场内的恒星以三角形的方式组合起来,提取星图模式,建立完备的训练样本集,使星图特征具有平移和旋转不变性。然后,采用BP神经网络识别星图特征,以权值矩阵代替导航星库,一旦网络训练完成,可以很快获得当前星图信息,实现星敏感器星图识别算法的实时性和鲁棒性;为了优化BP神经网络改进其自身缺点,采用PSO(粒子群算法)训练BP神经网络,获取使BP神经网络趋近全局最优的初始权值和阈值,使其加快收敛至全局最优。由实验结果表明,该星图识别算法识别率达100%。
In order to achieve accurate measurement of the current attitude of the spacecraft by the star sensor, how to improve the real-time performance and robustness of the star-map recognition algorithm becomes the key technology of the star sensor. This paper studies the pattern extraction applied in the process of star chart recognition, the establishment of training sample sets and the improvement of neural network training methods. First, we design a triangulation method based on the characteristics of the star map, combine the stars in the field of view in a triangular way, extract the star pattern, and establish a complete set of training samples to make the star map feature translation and rotation invariance . Then, the BP neural network is used to recognize the features of the star image, and the weight matrix is used to replace the navigation star database. Once the network training is completed, the current star image information can be obtained quickly to realize the real-time and robustness of the star sensor recognition algorithm. In order to optimize the BP neural network to improve its own shortcomings, PSO (Particle Swarm Optimization) is used to train the BP neural network to obtain the initial optimal weights and thresholds which approach the global optimality of the BP neural network so as to accelerate the convergence to the global optimum. The experimental results show that the recognition rate of this algorithm is 100%.