INTELLIGENT SYSTEM FOR CORRECTING GLOVE PATTERNS BASED ON MACHINE LEARNING BASED ON THE RESULTS OF FITTINGS

Abstract

The article presents a scientifically based approach to automating the correction of glove patterns based on machine learning methods and analysis of the results of fittings. The problem of exact matching the geometry of the pattern with the actual fit of the product remains one of the most difficult in sewing production, since traditional technologies are based mainly on manual measurements, expert assessments and empirical rules that do not allow reproducing local deformations with high accuracy and make it impossible to fully digitally automate the process of pattern correction. In this context, it is relevant to create a mathematical model capable of interpreting the visual data of fittings and predicting updated parameters of the glove contour, which correspond to the individual anatomical features of the user's hand.

The purpose of the study is to develop an intelligent system that combines classification and regression modules of a deep neural network to simultaneously determine the landing state and predict local geometric parameters necessary for pattern correction. The proposed model presents the shape of the pattern as a parameterized contour, and treats the landing deviation as local deformations that can be determined through the normalized coordinates of the object in the image. Thus, the system makes the transition from heuristic manual procedures to a formally defined machine prediction process.

For the first time, a two-component approach to visual data processing is integrated in the work: the classification unit determines the presence or absence of a glove on the hand, while the regression module predicts metrically interpreted correction parameters. The proposed methodology was implemented in a software package that covers the stages of data pre-processing, model training, comparative analysis with classical algorithms and visual-analytical interpretation of results. For validation, an open dataset of YOLO markup was used, containing images of a hand with and without a glove, provided with coordinates of local geometric fragments, which act as the basis for training the model.

The results obtained confirm the ability of the model to provide high classification accuracy and stable quality of forecasting local parameters, which allows them to be used as a basis for automated pattern correction. The scientific novelty of the work lies in the formalization of the process of pattern correction through the parameterized representation of deformations and the use of a hybrid neural network architecture. The practical significance lies in the possibility of integrating the system into the digital design process, reducing the number of fittings, reducing production costs and increasing the accuracy of the product. The developed approach can be extended to other garments and industries that require accurate modeling of local deformations.