Abstract:

Objective Long hours at desks and on phones are worsening human postures, leading to hunched backs and rounded shoulders. These changes directly harm the fit of a suit, affecting its overall look. Therefore, a characterization analysis of young female back shapes related to unhealthy body postures was prosposed, aiming to offer theoretical support for designing better-fitting tailored garments.

Method 3D body data were acquired using scanners, which were processed with Geomagic software for noise reduction and data filling. The back region was defined by dividing the body along the mid-sagittal plane and then characterized its morphology by extracting key feature points based on surface curvature. The data were the classified using gray correlation and the silhouette coefficient. Ultimately, a predictive model for back body shapes was established using a neural network, directly informing the optimization of shoulder pad designs in warp-knitted suits for different back types.

Results A study of 104 young women aged 18-25 defined the posterior back region as the area between the mandibular point and the horizontal circumference through the anterior abdominal protuberance. Using Geomagic reverse engineering software, the cross-sectional curves of the back region were generated. Local curvatures along these curves were calculated to locate the most prominent points, establishing nine landmarks. Based on these points, five planes were constructed to characterize back morphology. Multidimensional parameters involving six angles and three lengths were extracted from these feature points and planes.

The grey relational analysis (GRA) was adopted to extract the five key indicators from the feature parameters before K-means clustering was performed by randomly grouping these five indicators, where the cluster numbers (K) ranged from 2 to 9. For each test, the silhouette coefficient was calculated. The best result, with the highest coefficient, emerged when dividing subjects into three categories based on thickness and back-shoulder width. The first category was the straight and slender type, with an overall straight back representing the normal form. The second category was the broad-shouldered and thick-backed type, featuring a thick upper back, a broad back, and an externally rotated shoulder girdle, resembling winged scapulae. The third was the narrow-shouldered with posterior curvature type, presenting a prominent neck curve, narrow shoulders, and a mild kyphotic tendency.

A three-layer neural network was built in MatLab R2023b to predict the back body types of young women. Using a stratified sampling method, the dataset was divided into 3 groups, 70% for training, 15% for validation, and he rest 15% for testing, where the validation set was adopted to guide the training and to prevent overfitting, whereas the test set assessed the model's ability to generalize. The final model achieved an overall accuracy of 98.97%. Based on these distinct back shape classifications, the study concluded by applying optimized shoulder pad designs in warp-knitted suits tailored to each back type.

Conclusion Based on comprehensive 3D body scanning data, a systematic framework for back morphology analysis was established by defining precise anatomical boundaries and creating representative feature points and planes. Through advanced clustering methodology incorporating characteristic angles, lengths, gray correlation analysis, and silhouette coefficients, three distinct somatotypes was identified: the straight-slender variant representing standard morphology, the broad-thick type exhibiting winged scapulae characteristics, and the narrow-convex form demonstrating mild kyphotic predisposition. An accurate neural network prediction model was developed, achieving 98.97% classification accuracy, providing reliable technological support for customized shoulder pad engineering. Furthermore, significant correlations between scapular plane configuration and body thickness dimensions was revealed. Tailored shoulder pads designed for specific somatotypes effectively reposition the shoulder point posteriorly, creating optimized shoulder contours that reduce apparent body thickness. This strategic modification decreases critical distances between shoulder pads and back protrusion points, facilitating natural scapular retraction toward prominent areas.

These structural improvements collectively enhance shoulder silhouette definition, optimize overall body proportions, and elevate garment fit and aesthetic quality. The findings establish substantial theoretical and practical foundations for personalized apparel design and manufacturing processes.

Key words: apparel design and engineering, back shape, three-dimensional human body scanning, neural network, shoulder pad