基于服务机器人听觉的个体膳食构成自主感知算法

Autonomous Individual Dietary Composition Perception Algorithm Based on Social Robot Audition

针对现有服务机器人无法自主感知用户膳食构成的问题,提出了基于服务机器人听觉的个体膳食构成自主感知算法(AIDCPA)。首先,利用了基于梅尔频率倒谱系数和矢量量化算法的声纹识别方法识别出了说话人身份,并运用中文分词、词性标注和依存句法分析工具获取语音识别所得文本数据的语言特征;然后,给出了基于一阶谓词逻辑理论的推理定义和描述,提出了饮食组成获取的推理算法,进而形成了基于服务机器人听觉的个体膳食构成自主感知算法。为了评估AIDCPA算法性能,在服务机器人平台中实现了AIDCPA算法,并构建了训练和测试数据集。测试结果表明:系统感知饮食组成的F1值、精确度与召回率的均值分别为0.9491,0.9679和0.9407,有较强的感知鲁棒性。

Focusing on the problem that current social robots are not able to perceive user’s dietary composition autonomously,this paper proposed AIDCPA,an autonomous individual dietary composition perception algorithm based on social robot audition. Firstly,a voiceprint recognition method based on Mel Frequency Cepstrum Coefficient( MFCC) and vector quantization algorithm was used to identify speaker’ s identity,and then the Chinese word segmentation,part-of-speech tagging and dependent parsing tools were employed to deal with the text data captured from the audition to obtain its linguistic features. Then,the definition and description of reasoning based on the first-order predicate logic theory was proposed,and the reasoning algorithm of dietary composition perception was proposed,and then the AIDCPA was designed. To evaluate the performance of AIDCPA,the proposed algorithm in a social robot platform was implemented,and two datasets including training dataset and test dataset were prepared. Extensive dietary composition perception experiments on the social robot show that AIDCPA is capable of perceiving dietary composition with an average F1 score,precision rate and recall rate of 0.9491,0.9679 and 0.9407,respectively,indicating the good robustness of AIDCPA algorithm.