基于卷积神经网络的服务机器人听觉隐私信息分类算法  

Social Robot Auditory Privacy Information Classification Algorithm Based on Convolutional Neural Network

作　　者：王怀豹 杨观赐[1] 李杨[1] 林家丞 WANG Huaibao;YANG Guanci;LI Yang;LIN Jiacheng(Key Laboratory of Advanced Manufacturing Technology of Ministry of Education,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学现代制造技术教育部重点实验室,贵州贵阳550025

出　　处：《贵州大学学报（自然科学版）》2020年第3期76-80,共5页Journal of Guizhou University:Natural Sciences

基　　金：国家自然科学基金项目资助(61863005);贵州省科技计划项目资助(黔科合平台人才[2018]5702,黔科合平台人才[2018]5781,黔科合平台人才[2020]6007,黔科合支撑[2019]2814,黔科合支撑[2020]4Y056)。

摘　　要：为获得机器人听觉行为隐私感知方法,解决语音监听设备存在的隐私泄露风险,本文提出了基于卷积神经网络的服务机器人听觉隐私信息分类算法(APICA)。首先,设计了基于卷积神经网络的服务机器人听觉隐私信息分类算法及其卷积神经网络模型;其次,给出了机器人的听觉隐私信息监听系统工作流程;最后,为评估该听觉隐私信息分类算法性能,构建了训练和测试数据集,并在服务机器人平台上部署和实现了该算法。测试结果表明:系统识别隐私信息的平均精确率P、召回率R和F1值分别为96.35%、93.20%和94.53%,具有良好的识别和分类效果。In order to obtain the robot auditory behavior privacy perception method and solve the risk of privacy leakage in speech monitoring equipment,APICA,a social robot auditory privacy information classification algorithm based on Convolutional Neural Network was proposed.Firstly,the algorithm and convolutional neural network structure of the proposed APICA were presented,and then the work flow of the social robot auditory privacy information monitoring system was put forward.Finally,to evaluate the performance of APICA,training dataset and test dataset were established,and the proposed method was implemented in a social robot platform.The test results show that APICA is capable of recognizing privacy information with an average accuracy rate,recall rate and F1 score of 96.35%,93.20%and 94.53%,respectively.

关 键 词：隐私感知方法 服务机器人 分类算法 听觉隐私信息 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]