基于嵌入式智能平台的目标检测算法设计  

Detection Algorithm Designing Based on Embedded Intelligent Platform

作　　者：王开 曹兴钰 徐锦涛 WANG Kai;CAO Xin-yu;XU Jin-tao(Xi′an Aeronautics Computing Technique Research Institute,AVIC,Xi′an 710000,China)

机构地区：[1]航空工业西安航空计算技术研究所,陕西西安710000

出　　处：《航空计算技术》2024年第5期94-98,共5页Aeronautical Computing Technique

基　　金：航空科学基金项目资助(2022Z071031003)。

摘　　要：近年来,随着深度学习算法的发展,设计适用于嵌入式智能计算平台的目标检测算法需求也愈发紧迫。介绍了嵌入式智能计算平台下目标检测算法的设计准则。详细分析嵌入式平台的计算能力和资源限制,确定算法设计的约束条件;根据目标检测任务的需求,选择合适的深度学习算法为基础,并基于NPU硬件特征优化模型,从而使得应用算法在嵌入式平台性能更优。此外,基于昇腾310为核心的嵌入式智能计算平台,提出针对特定计算密集型算子的轻量化方法,从而提高算法的实时性。实验结果表明,所提方法在嵌入式平台上可实现较高检测精度和较低延迟,满足了具体场景应用的需求,并为嵌入式智能计算平台下的目标检测算法设计提供参考。Recent years,with the development of deep learning,the need for designing detection algorithms suitable for embedded intelligent computing platforms has become increasingly urgent.This article introduces the design criteria for detection algorithms under embedded intelligent computing platforms.Firstly,a detailed analysis of the computing capabilities and resource limitations of the embedded platform is conducted to determine the constraints for algorithm design.Secondly,based on the requirements of the detection task,we should select an appropriate deep learning algorithm as the foundation and optimize the model based on the hardware characteristics of the NPU(Neural Processing Unit),so that the application algorithm can achieve better performance on the embedded platform.In addition,based on the embedded intelligent computing platform with Ascend 310 as the core,this article proposes optimization methods for specific computationally intensive operators,thereby improving the real time performance of the algorithm on this hardware.Experimental results show that the method proposed in this article can achieve dual optimization of accuracy and latency on embedded platforms,satisfying the needs of specific scenario applications and providing a reference for the designing of detection algorithms under embedded intelligent computing platforms.

关 键 词：嵌入式平台 智能计算 目标检测 算法设计 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]