基于注意力模型的多模态特征融合雷达知识推荐  

A multi-modal feature fusion radar knowledge recommendation method based on attention mode

作　　者：李稳安[1] 陈柳柳 陈实 LI Wen’an;CHEN Liuliu;CHEN Shi(Nanjing Research Institute of Electronic Technology,Nanjing 210039,P.R.China;School of Business,Emporia State University,Kansas,the United States)

机构地区：[1]南京电子技术研究所,南京210039 [2]恩波利亚州立大学商学院,堪萨斯美国

出　　处：《重庆大学学报》2021年第7期34-42,共9页Journal of Chongqing University

基　　金：国防基础科研资助项目(A1120131044)。

摘　　要：为了能够在数量庞大的雷达技术资料中快速准确地找到科研人员感兴趣的雷达知识信息并进行推荐,提出了一种基于注意力模型的多模态特征融合雷达知识推荐方法,学习高层次的雷达知识的多模态融合特征表示,进而实现雷达知识推荐。该方法主要包括数据预处理、多模态特征提取、多模态特征融合和雷达知识推荐4个阶段。实验结果表明:与只利用单一模态特征以及简单串联多模态特征的方法相比,利用文中方法学习到的多模态融合特征进行雷达知识推荐,推荐结果的准确率、召回率和综合评价指标(F_(1)值)均有显著提高,表明提出的基于注意力模型的多模态特征融合方法对于知识推荐任务更加有效,体现了算法的优越性。In order to quickly and accurately find the radar knowledge and information that researchers are interested in and make recommendations from a large number of radar knowledge data,this paper proposes a multi-modal feature fusion radar knowledge recommendation method based on attention model,which can learn high-level fusion feature representation from multiple modalities for the radar knowledge.Then the radar knowledge recommendation can be performed using the learned fusion feature.The proposed method consists of four stages,i.e.,data preprocessing,multi-modal feature extraction,multi-modal feature fusion,and radar knowledge recommendation.Extensive experiments demonstrate that,compared to the existing methods using only single-modal features and simple concatenation of multi-modal features,the multi-modal fusion features learned by the proposed method achieves significant improvement in precision,recall,and F_(1) value for radar knowledge recommendation,suggesting its effectiveness for radar knowledge recommendation.

关 键 词：雷达知识推荐 多模态 特征融合 深度神经网络 注意力模型 

分 类 号：TP520.20[自动化与计算机技术]