摘要
智慧城市管理模式的提出,意味着基于深度学习的目标识别技术将在城市发展中起到重要作用。为提高识别技术的精确度,提出一种利用RNN的空间结构特征提取方法,利用CUDA的并行加速器来提高实时性能,将提取器耦合进目标特征提取基础网络模型,形成多融合特征的目标检测网络,最后提出一种快速排序算法来加快网络的整体运行速度。检测结果显示:多融合特征目标检测网络相较于其他两个网络而言,只有当置信度足够高时,目标识别的召回率才下降到0.5以下。目标识别技术的改进扩大了视觉技术的应用范围,有利于促进智慧城市建设。
Smart city management mode means that target recognition technology based on deep learning will play an important role in urban development.In order to improve the accuracy of recognition technology,a spatial structure feature extraction method using RNN is proposed,in which CUDA′s parallel accelerator is used to improve the real-time performance,and then the extractor is coupled into the basic network model of target feature extraction to form a target detection network with multiple fusion features.Finally,a fast sorting algorithm is proposed to speed up the overall operation of the network.The detection results show that compared with the other two networks,only when the confidence is high enough,the recall rate of target recognition will drop to below 0.5.The improvement of target recognition technology expands the application scope of vision technology,which is conducive to promoting the construction of smart city.
作者
胡琼
HU Qiong(Anhui Lu′an vocational and technical college,Lu′an 237158,Anhui,China)
出处
《贵阳学院学报（自然科学版）》
2021年第2期30-34,共5页
Journal of Guiyang University：Natural Sciences
基金
2019年安徽省高校优秀青年人才支持计划项目“基于深度学习的智慧城市关键目标识别研究”(项目编号:gxyq2019206)。
关键词
智慧城市管理
深度学习
目标识别
特征提取
Smart city management
Deep learning
Target recognition
Feature extraction
