摘要
本文以卷积神经网络为基础,结合传统的图像检索算法VLAD对其进行扩展,并对室外场景的识别任务开展了室外场景特征和场景识别神经网络的构建与实现。同时,通过利用室外场景数据集,对研究成果进行了室外场景识别预测实验,利用特征提取实验探究了网络学习内容。实验结果表明,相较传统方法,本文设计方案在Tokyo24/7数据集上能达到90%的召回率和88.51%的查询精度,有效克服了室外场景识别中光照、拍摄角度与变化因素(行人与车辆)带来的识别难题。
Based on convolutional neural network,this paper extends the traditional image retrieval algorithm VLAD,and carries out the construction and implementation of outdoor scene features and scene recognition neural network for outdoor scene recognition tasks.At the same time,outdoor scene recognition and prediction experiments were conducted on the research results using outdoor scene datasets,and feature extraction experiments were used to explore the content of online learning.The experimental results show that compared to traditional methods,the design scheme in this paper can achieve a recall rate of 90%and a query accuracy of 88.51%on the Tokyo 24/7 dataset,overcoming the recognition difficulties caused by lighting,shooting angle,and changing factors(pedestrians and vehicles)in outdoor scene recognition to a certain extent.
作者
王亚军
穆杞梓
余末银
王洪海
朱立远
WANG Yajun;MU Qizi;YU Moyin;WANG Honghai;ZHU Liyuan(CHN Energy Digtal Intelligence Technology Development(Beijing)Co.,Ltd.,Beijing 100011,China;State Key Laboratory of Information Engineering in Surveying,Mapping and Remote Sensing,Wuhan University,Wuhan,Hubei 430079,China)
出处
《自动化应用》
2023年第14期201-207,共7页
Automation Application
关键词
卷积神经网络
室外场景识别
VLAD算法
图像检索
特征编码
convolutional neural network
outdoor place recognition
VLAD algorithm
image retrieval
feature encoding
