Fig. 5From: EKF-based self-attitude estimation with DNN learning landscape informationProposed network architecture. It consists of CNN layers and FC layers. The input data is a resized image, and the output data are a mean vector and a covariance matrix. They are computed with an output from the final layer as Eqs. 2, 3, respectively. Log-probability of multivariate normal distribution is used as a loss function of this modelBack to article page