Search

In this paper, an end-to-end deep learning framework for reconstructing 3D models by computing depth maps from multiple views is proposed. An unsupervised 3D reconstruction method based on multi-view propagation is introduced, which addresses the issues of large GPU memory consumption caused by most current research methods using 3D convolution for 3D cost volume regularization and regression to obtain the initial depth map, as well as the difficulty in obtaining true depth values in supervised methods due to device limitations. The method is inspired by the Patchmatch algorithm, and the depth is divided into n layers within the depth range to obtain depth hypotheses through multi-view propagation. What's more, a multi-metric loss function is constructed based on luminosity consistency, structural similarity, and depth smoothness between multiple views to serve as a supervisory signal for learning depth predictions in the network. The experimental results show our proposed method has a very competitive performance and generalization on the DTU, Tanks & Temples and our self-made dataset; Specifically, it is at least 1.7 times faster and requires more than 75% less memory than the method that utilizes 3D cost volume regularization.

Summary: The control problem of the hypersonic vehicles is studied in this article. A new control approach is presented. This approach consists of a data‐driven dynamic model established by multiple neural networks, an online identification method for system parameters, and a basic backstepping controller. The implementation of this approach requires a dynamic model and system parameters including the moment of inertia and aerodynamic parameters of the hypersonic vehicles. The parameter identification problem is regarded as a dynamic optimization process. The loss function is designed by the Lagrange criterion, and its constraints are determined by the physical and the numerical values. In the case of model mutation, the system parameters identified online are used as the nominal values of the output of the neural network in the data‐driven model to adjust the controller through its gradient descent. Simulation comparisons are given to show the effectiveness of the proposed data‐driven approach. [ABSTRACT FROM AUTHOR]

This paper addresses a difficult problem of strength allocation strategies generation with various adversaries and complex factors. Firstly, to investigate the strength allocation strategies generation problem, an attack‐defense differential game problem is formulated based on an improved Lanchester equation. Secondly, a numerical method, multi‐intervals simultaneous orthogonal collocation decomposition (MISOCD) method, is proposed to obtain the strength allocation strategies from the constructed model. Compared with the analytical method, MISOCD does not need to derive the necessary conditions. Thirdly, this study designs an approximated solution generation strategy based on adaptive learning pigeon‐inspired optimization algorithm to pregenerate the approximated strength allocation strategies in order to solve the initial value sensitivity problem. The approximated strategies are then used as the initial value guess of MISOCD method to generate optimal strength allocation strategies. Finally, two attack‐defense numerical simulations verify the effectiveness of strength allocation strategies generated by the proposed approach. Our proposed results provide a theoretical guide for both making attack‐defense strength allocation strategies and assessing confrontation actions. [ABSTRACT FROM AUTHOR]