摘要
针对传统数字孪生猪场建模过程中存在的硬件依赖性强、成本高昂、建模效率低等问题,基于生成式人工智能技术(简称AIGC)提出了一种数字孪生猪场快速建模方法。通过分析多视角重建、结构光扫描、激光扫描等传统建模技术的特点与局限性,结合前期手工建模积累的经验,从硬件需求、重建质量、实时性能等维度与该方法进行了对比。在Jetson Nano平台上对优化的Shap-E模型进行了验证实验,以突破边缘计算平台的硬件约束,并与TRELLIS方法进行了对比分析。仿真实验表明,优化后的Shap-E模型在处理连续帧数据时约有2.3 s/帧的延迟,静态建模能满足养殖场个体识别(准确率82.4%)和体态评估(误差率<15%)等基础需求,显著优于其他建模方法。
Aiming at the problems of strong hardware dependence,high cost,and low modeling efficiency in traditional pig digital twins modeling methods,a fast modeling method for pig farm digital twins based on generative artificial intelligence technology(AIGC)is proposed.By analyzing the characteristics and limitations of traditional modeling techniques such as multi-view reconstruction,structured light scanning,and laser scanning,and combining the experience accumulated from previous manual modeling,the application advantages of AIGC technology were compared from the dimensions of hardware requirements,reconstruction quality,and real-time performance.The optimized Shap-E model was verified on the Jetson Nano platform to break through the hardware constraints of the edge computing platform,and was compared with the TRELLIS method.Simulation experiments show that the optimized Shap-E model has a delay of about 2.3 s/frame when processing continuous frame data.Static modeling can meet the basic requirements of individual recognition(accuracy 82.4%)and posture evaluation(error rate<15%)in pig farms,and is significantly better than other modeling methods.
作者
杨俊
刘波
YANG Jun;LIU Bo(College of Information and Intelligence,Hunan Agricultural University,Changsha,Hunan 410128,China)
出处
《农业工程与装备》
2024年第5期27-31,34,共6页
AGRICULTURAL ENGINEERING AND EQUIPMENT
基金
国家自然科学基金项目(61972147)
云南省重大科技专项计划(202202AE090032)
湖南省大学生创新创业训练项目(s202410537132、s202410537131、s202410537126)。
关键词
数字孪生
生成式人工智能:猪场建模
digital twin
generative artificial intelligence:modeling of pig farms
