基于深度学习的非合作航天器姿态估计

针对空间非合作航天器姿态测量时受光照和地球背景影响大的问题,提出了一种基于卷积神经网络的端到端姿态估计方法.在该方法中,主干网络采用AlexNet与ResNet.首先,移除主干网络末端的全连接层,并列连接3个全连接层,采用三分支网络分别对姿态角进行估计.然后,设计了将分类问题与回归问题相结合的损失函数,通过分类方法将姿态估计限定在某一范围内,再使用回归方法进一步微调姿态.姿态分类损失函数确定姿态角度基准点,姿态回归损失函数对估计角度进行微调.相较于仅采用回归方法进行姿态估计,此方法能够有效减小姿态估计平均绝对误差、标准差与最大误差.实验对比了不同主干网络的测量精度,平均绝对误差在0.376°～0.746°之间,最优标准差为0.474°.

Attitude Estimation of Non-cooperative Spacecraft Based on Deep Learning

Aiming at the problem that the attitude measurement of non-cooperative spacecraft is greatly affected by illumination and Earth background, an end-to-end attitude estimation method based on Convolutional Neural Network(CNN) is proposed for attitude measurement of non-cooperative targets. In this method, AlexNet and ResNet are used as the backbone network. Firstly, the full connection layer at the end of the backbone network is removed, and three full connection layers are connected in parallel. The attitude angle is estimated by three branch networks. Secondly, the loss function which combines the classification problem and the regression problem is designed. The attitude estimation is limited to a certain range by the classification method, and then the regression method is used to further fine tune the attitude. The attitude classification loss function determines the datum point of the attitude angle, and the attitude regression loss function adjusts the estimated angle. Compared with the regression method, this method can effectively reduce the average absolute error, standard deviation and maximum error of attitude estimation. The measurement accuracy of different backbone networks is compared, the average absolute error ranges from 0.376 ° to 0.746 ° and the optimal standard deviation is 0.474°.