Search

Heterogeneous multi-processor architecture which achieves rich functionalities with different types of processors, is widely used to provide powerful calculating capability while keeping energy consumption under control. Although this architecture can improve system flexibility for adapting to future requirement variations, it results in a complex multi-task scheduling problem for system designers to produce a reasonable schedule that satisfies all deadline, data dependency, and energy limitation constraints. In this paper, we concentrate on the energy consumption-constrained scheduling problem of workflows in heterogeneous multi-processor embedded systems. First, we model the workflows and energy consumption of processors, and formulate the energy consumption-constrained scheduling problem as an optimization one whose objective is to shorten the schedule length of workflows as much as possible. Then, with an improved energy per-assignment strategy, we propose a novel energy difference coefficient-based scheduling algorithm to produce an approximately optimal allocation of processors, frequencies, and start times for each task while guaranteeing that the data dependency and energy limitation constraints are satisfied. Finally, experiments on both randomly-generated and real-world workflows are conducted to verify the reliability and efficiency of the proposed approach. [ABSTRACT FROM AUTHOR]

Abstract: The recent spectacular progress in modern microelectronics created a big stimulus towards development of embedded systems. Unfortunately, it also introduced unusual complexity which results in many serious issues that cannot be resolved without new more adequate development methods and electronic design automation tools for the system-level design. This paper discusses the problem of an efficient model-based multi-objective optimal architecture synthesis for complex hard real-time embedded systems, when using as an example a system-level architecture exploration and synthesis method that we developed. [Copyright &y& Elsevier]

Many of the broadcasting requirements, setting beach motion data from a dedicated camera and extracting meaningful information from the data raise important research agendas. Therefore, computer vision and machine learning are essential for the automatic or semiautomatic processing of heterogeneous multiprocessors moving to the beach. The detection accuracy of moving images is low, which takes more time. The proposed system has emerged to identify and classify beach sports services from the deployment of arm-worn gyroscopes for semi-professional beach sportspeople. It is based on off-the-shelf heterogeneous multiprocessors of convolutional neural networks that can distinguish between common service types. This shows the potential of wearable technology for beach sports that can provide accurate motion analysis. Preprocessing is used to remove noisy and irrelevant data. Segmentation is the process of splitting a moving image into components or objects in the image. Classification analysis, and offer heterogeneous multi-processor analyzes the latest detailed performance evaluation based on Convolutional Neural Network system. Therefore, this study will use Heterogeneous Multi-Processor beach sports as raw material image detection technology analysis. In this study, an equal edge detection image detection technique, gradation processing, target acquisition and target recognition to the sports video's actual needs combined to meet the needs of different dynamic image detection. [ABSTRACT FROM AUTHOR]