Earticle

초록

영어

This study continues the entire process of researching how to build a pet digital twin, and continues to draw conclusions through cluster surveys and analysis of experimental results. First, with the advancement of AI models, the Transformer model achieved 98.5% accuracy in processing time-series sensor data, demonstrating superior performance compared to existing CNN-LSTM (Convolutional Neural Network-Long Short-Term Memory) (95.8%) or RF (Random Forest) (91.2%). Second, to overcome the differences in body shape between dog breeds in vision-based pose estimation using multimodal and shape-adaptive technologies, a family-aware model such as AniMer is essential, and accuracy is secured through fusion with an IMU (Inertial Measurement Unit) sensor. Third, the architecture, which applies its own framework and the MQTT (Message Queuing Telemetry Transport) protocol, a cloud-centric lightweight protocol, through platform standardization, enables real-time synchronization between physical sensor data and virtual engines (Unity/Unreal Engine), and provides the foundation for a bio-digital twin that visualizes even invisible information such as bioenergy through software sensors. Beyond a simple technological endeavor, companion animal digital twins will become established as a “digital companion” technology that expands communication channels with companion animals who cannot speak and dramatically improves their quality of life.

중국어

本研究全面探讨了面向宠物数字孪生构建的行为模式研究全流程,通过文献调研与实验结 果分析,得出如下结论.第一,随着人工智能模型的优化升级,在时间序列传感器数据处理任务 中,基于Transformer架构的模型达到了98.5%的准确率,相较传统的卷积长短期记忆网络 (95.8%) 和随机森林算法(91.2%) 展现出更优性能,这表明自注意力机制在捕捉宠物犬行为 的长期关联特征方面具有显著效果.第二,在基于计算机视觉的姿态识别任务中,借助多模态与 形态自适应技术,想要克服不同犬种之间的体型差异,AniMer这类犬种识别适配模型不可或缺 ; 同时,通过与惯性测量单元传感器的数据融合,进一步保障了识别精度.第三,在平台标准 化层面,集成自研框架与消息队列遥测传输协议的系统架构,实现了物理传感器数据与虚拟引擎 (Unity/虚幻引擎) 的实时同步,依托软件传感器完成了生物能量等非可视化信息的可视化呈 现,为构建宠物生理数字孪生奠定了技术基础.宠物数字孪生的价值远超单纯的技术尝试,它将 作为一种“数字伙伴”技术,拓宽与无法言语的宠物犬之间的沟通渠道,显著提升宠物的生活质 量.未来的研究方向,有望朝着能够实时学习个体特性并持续进化的“自适应数字孪生”方向发 展,同时推动其与低延迟通信􀕵边缘计算技术的深度融合.

목차

Abstract
I. 서론
1. 연구 배경 및 필요성
2. 연구 목적
II. 이론적 배경
1. 동물 행동학(Ethology)과 디지털 행동목록(에토그램, Ethogram)
2. 웨어러블 센서 및 관성 측정 장치(IMU, Inertial Measurement Unit) 기반 연구
3. 컴퓨터 비전 및 포즈 추정(Pose Estimation)
III. 연구 방법
1. 멀티모달 데이터 파이프라인(Multimodal data pipelines) 구축
IV. 연구 결과 및 고찰
1. 실험 모델 아키텍처(Architecture)
2. 실험 결과 및 성능 지표 비교
3. 결과 분석
4. 컴퓨터비전(Computer Vision) 모델에서 품종 일반화의 난제 및 해결책
5. 디지털 트윈 시스템 아키텍처 설계
6. 고찰
V. 결론
참고문헌
中文摘要

키워드

저자

• 김수영 [ Su-Young Kim | (주)소프트젠 ]
• 김수진 [ Su-Jin Kim | 동명대학교 반려동물보건학과 교수 ] Corresponding Author

참고문헌

발행기관

간행물

표지보기 관심저널 등록

• 간행물명

  국제보건미용학회지 [Journal of Health and Beauty]

• 간기

  계간

• pISSN

  2005-369X

• eISSN

  2765-1347

• 수록기간

  2007~2026

• 등재여부

  KCI 등재

• 십진분류

  KDC 593 DDC 646