Research On Garment Patterns Classification Based On Graph Neural Network

Current production in the clothing industry,cutting and sewing processes have reached a certain degree of automation,while garment pattern making is highly dependent on the experience of designers and pattern makers,and its intelligent research is still in its infancy.The automatic classification of garment patterns is an important technology to realize the intelligentization of clothing plate making.Garment patterns are engineering drawings for product layout,cutting and sewing,complexity is reflected in:(1)the type of multi-entity,comprising points,line segments,circles,arcs,splines,etc.;(2)entities organization complex,multi-level hierarchical structure,entities coupled with each other;(3)the placement of garment components(e.g.,collar,pocket,etc.)in the model diagram is arbitrary;(4)no uniform standard for the graphic element expression used in the template diagram(e.g.,line width,color etc.)This paper proposes two methods of transforming garment pattern into graph.They are methods based on centroid and curve equations and method based on nonuniform rational B-spline control points,further propose a graph convolutional neural network GPC-GCN(Garment Pattern Classification Graph Convolutional Network)that can process these two types of graph.On the linear fractal dataset(13500 samples)and the actual garment pattern dataset(about 161,000 samples),experiments were carried out on the GPC-GCN proposed in this paper.The results show that:(1)The proposed graph data modeling method can not only maintain the shape information of each component in the garment pattern,but also solve the problem of arbitrary placement of garment components;(2)Compared with the existing image-based convolutional neural network and graph neural network,the proposed graph data modeling method and graph neural network GPC-GCN have better performance in the classification of garment patterns.