卷 45, 编号 6 (2019)

A method of labeling aerial and satellite remote sensing images for detecting objects of a priori known shapes is proposed. The implementation of this method uses an algorithm of piecewise linear approximations of the boundary of template objects and an algorithm of transforming the template contour descriptions to fit them to images of objects in the photographs. The developed algorithms are tested using the practical problem of determining the types of aircraft in remote sensing images remote sensing images. It is shown that the proposed method improves the labeling accuracy and is faster than the manual detection of aircraft images in photographs.

Management of access to the Input/Output (IO) facilities of a computer system is one of the core functions of the operating system kernel. In this paper, we propose a new design of port-mapped IO management subsystem, suitable for use in the second generation or true microkernels. Specifically, the proposed design uses IA-32, the widespread CPU with the support of the concept of IO ports, as a target processor architecture. We also present and describe the protocol for access to IO means implemented in the kernel from the user-mode drivers, as well as the related optimizations implemented on the kernel side.

The complex organization and application conditions of space systems for automatic identification and localization of mobile objects, which include the automatic identification system (AIS) and automatic dependent surveillance-broadcast (ADS-B) system, determine the choice of simulation models for the mathematical formalization of their operation. Simulation modeling of satellite constellations capable of receiving, processing, and retransmitting AIS and ADS-B signals can take a significant amount of time when being used to substantiate circuit design solutions for satellites and plans for their further application given a large number of radiation sources to be simulated (e.g., for the AIS, their number exceeds 500 thousand). One of the methods for solving this problem is parallel computing based on the compute unified device architecture (CUDA) technology. However, due to the specificity of machine instruction execution on NVIDIA GPUs, software quality depends heavily on GPU memory allocation efficiency and algorithms for program code execution. In this paper, we propose a method for target performance evaluation of space systems for automatic identification and localization of mobile objects; the method uses massively parallel computations on GPUs to provide a significant reduction in simulation time, which is especially important for multi-satellite constellations. The efficiency of the method is confirmed by model-cybernetic experiments carried out on various software and hardware platforms.