基于思维进化算法优化的女裤合体性智能预测模型构建与应用OA

When purchasing pants online,consumers are unable to try them on,making it difficult to judge fit.This is a key reason for the high return rate of women's clothing,which not only affects the consumer experience but also increases e-commerce operating costs.Existing solutions rely on try-on,involve complex processes,or fail to accurately depict the nonlinear relationship between human body and clothing size,making it difficult to adapt to large-scale online scenarios. To overcome these limitations,this study aims to develop a non-contact,automated,high-precision fit prediction model for women's pants.The core objective is to achieve precise fit-level predictions without fitting sessions,using only readily available 2D human body images and key garment dimension data.This will provide a reliable fit prediction solution for online pant purchases,and reduce return rates accordingly.The study first employed a 3D scanner to capture body data from 35 young women.Using computer vision technology,it automatically extracted 10 key point distance features—including waist width,abdominal thickness,and hip width—along with three body shape descriptors from the generated 2D front and side profile images.Concurrently,garment features were defined as waist circumference,hip circumference,inseam length,and crotch width.On the Style3D virtual fitting platform,each virtual mannequin was matched with six different sizes of women's pants,generating 210 matching samples.Fit quality labels were assigned by professionals using a five-point rating scale,incorporating virtual pressure maps and visual appearance evaluations.After validating feature effectiveness,a prediction model was constructed using a mind evolution algorithm(MEA)to optimize the backpropagation(BP)neural network.This model leveraged MEA's global search capability to optimize the BP network's initial parameters,so as to overcome its tendency to get stuck in local optima. The research results show that the constructed MEA-BP model achieved a prediction accuracy of 90.1%for the fit of women's trousers,significantly higher than the 78.5%accuracy of the traditional BP neural network model.To further validate the model's generalization capability,15 new samples not involved in training were used for independent testing,achieving a prediction agreement rate of 86.67%for the MEA-BP model,which outperformed the traditional BP neural network,Bayesian algorithm,support vector machine model,and random forest model.These findings indicate that the human-body-and-clothing feature system extracted based on computer vision can effectively represent fit information;the BP neural network model optimized by MEA can learn and predict the fit of women's trousers more accurately,surpassing the performance of the traditional BP neural network.This method provides an objective,non-contact fit prediction tool for online clothing shopping,demonstrating practical application potential.