基于深度学习的多模态融合图像识别研究  

Research on multi-modal fusion image recognition based on deep learning

作　　者：秦放[1] 曾维佳[1] 罗佳伟[1] 徐鹏 QIN Fang;ZENG Wei-jia;LUO Jia-wei;XU Peng(Dalian University of Science and Technology,Dalian 116052,Liaoning Province,China)

机构地区：[1]大连科技学院,辽宁大连116052

出　　处：《信息技术》2022年第4期29-34,共6页Information Technology

基　　金：辽宁省教育厅2019年度科学研究经费项目(W2-019003)。

摘　　要：基于深度学习的方法,利用多模态信息融合技术,将汽车等目标的声音、图像信息融合,用于对测试目标的判断。利用改进Inception网络对图像识别技术展开研究,分别比较了单纯图像信息情况的辨识、单纯声音信息情况的辨识,以及多模态融合情况下的辨识。通过试验分析和比较,证明该方法可以有效提高用户的图像识别能力,将特种车辆的平均识别精确度提升到97%以上,可用于特种车辆紧急避让等情况,具有很好的应用价值。Based on the method of Deep Learning,this paper uses the multi-modal information fusion technology to fuse the sound and image information of the car and other targets for the judgment of the test target.The improved Inception network is used to study the image recognition technology.The identification of pure image information,pure sound information and multi-mode fusion are compared respectively.Through experiment analysis and comparison,it is proved that the method can effectively improve the user’s image recognition ability,and the average recognition accuracy of special vehicles can be increased to more than 97%,which can be used in special vehicle emergency avoidance and other situations,and has good application value.

关 键 词：深度学习 多模态融合 图像识别 声音识别 

分 类 号：TM715[电气工程—电力系统及自动化]