基于深度学习的针织全成形服装定制模型生成方法

The knitted garment customization model generation technology is a key step in the study and development for knitting garment customization and diversification, and the small scale, rapid response and flexible manufacturing process. It is always a hot research area in current garment digitalization technology. The full knitted garment is a type of garment which is formed into three-dimensional garment in one step by knitting the yarn with the full-forming knitting equipment. This short production process of garment has completely changed the traditional production mode of the fabric-cutting garment, and because of present structural model of traditional fabric-cutting garment is not suitable for the full knitted garment, the usage of deep learning to achieve the automatic generation of full knitted garment custom models has a great significance to promote the development of related garment industries. In this project: (1)A dataset of garment classified style and 2D/3D model is built for the full knitted garment modeling; (2)The style and structural features is extracted automatically through a convolutional neural network model trained with deep learning algorithms; (3)The automatic generation model of three-dimensional full knitted garment from two-dimensional garment image based on the graph convolutional neural network is studied and the three-dimensional virtual display of garment is also implemented. This research aims to provide scientific support for the development of online personalized custom and flexible manufacturing technology for full knitted garments, which has significant value for the theoretical research and practical application.

针织服装定制模型生成技术是针织服装定制与多样化、小规模、快速反应柔性制造研发的关键环节，是当前数字化服装技术的研究热点。针织全成形服装是由纱线通过针织全成形设备一次编织成三维立体的服装，一线成衣短流程生产彻底变革了面料裁剪类服装的传统生产模式，目前传统裁剪服装结构模型不适用于针织全成形服装，因此，运用深度学习算法实现针织全成形服装定制模型自动生成，对推动相关产业发展具有重要意义。本项目的主要研究内容包括：（1）建立针织全成形服装分类款式图及模型数据集；（2）基于深度学习算法训练深度卷积神经网络模型，自动提取针织服装图片款式及结构特征；（3）基于深度学习的图卷积神经网络，研究针织全成形服装从二维图片到三维服装网格模型的自动生成模型及实时三维虚拟展示。旨在为针织全成形服装在线个性定制与柔性制造技术发展提供科学支持，具有理论研究意义与实践应用价值。

随着云计算和人工智能技术的快速发展，服装数字化技术在服装的计算机辅助设计制造领域的研究是目前服装CAD技术的热点研究内容之一。本项目基于深度学习的针织全成形服装定制模型的生成技术，通过利用款式图和三维服装模型数据集，训练并生成基于卷积神经网络(CNN)和图卷积神经网络(GCN)结构的深度学习模型。模型从二维款式图中提取款式特征图，预测并实时生成与图片特征一致的三维针织全成形服装网格模型，解决服装三维模型生成速度缓慢的问题，采用数据驱动技术丰富和完善服装模型数据集，训练优化深度学习模型，预测服装形变和人体模型之间的特征映射关系，运用简化线性运算实现针织服装三维模型的实时编辑和虚拟展示功能。同时通过建立的网格映射关系，确定线圈结构在三维针织成形服装网格模型上的位置，利用C#和JS技术沿着特定的路径进行渲染，实现三维针织成形服装虚拟仿真。课题为针织全成形服装二维/三维实时交互式CAD设计定制系统的开发与应用提供理论模型和科学支持，具备良好的理论意义和应用前景。