Abstract:

Objective Traditional designs of head, neck and shoulder products ignore individual shape differences, causing user discomfort. Taking the young group in East China as the target, 3-D scanning technology and machine learning are used in this study to establish an intelligent classification system for head, neck and shoulder shapes. The research aims to provide a scientific basis for personalized engineering design (such as pillows) to enhance product comfort. The objectives of this study include identifying key morphological indicators, establishing intelligent classification models, and visualizing the results through 3-D modeling.

Method Data of 228 participants were collected using the Vitus Smart XXL 3-D scanner. Seven indicators were extracted from the model processed by Geomagic Studio, which are tragus point-lateral width of the acromion, lateral depth of the neck, wall-anterior shoulder distance, occipital process point distance, cervical depth, occipitocervical distance and scapular distance of the neck. The normality of the data was verified through the Kolmogorov-Smirnov and Shapiro-Wilk tests. Shape classification was carried out using the K-means clustering method of SPSS 27.0, and a 3-D visualization model was generated through 3ds Max software. The intelligent classification model was established by using computer ensemble learning.

Results By analyzing the 3-D human body scan data, no significant differences were identified in the head, neck and shoulder data between men and women, and the scanned data of men and women were combined for discussion. Using SPSS cluster analysis, the head, neck and shoulder shapes of 222 young people from East China were classified into four types, namely the normal type, deep neck type, low occipital flat neck type and high occipital deep neck type. All the indicators of the normal group were within the median range, manifesting that no significant morphological abnormalities were detected in the head, neck and shoulder structure. The occipital process distance of the low occipital and flat neck type group was found significantly lower than that of the normal type (with a mean difference of 3.5 cm), indicating that the position of the posterior occipital point shifted relatively backward, the cervical depth distance shortened by 1.6 cm, and there was an abnormal physiological curvature of the cervical vertebrae. The depth of the neck in the deep-necked group was 1.3 cm smaller compared with the normal type. The group with high occipital and deep neck type showed typical characteristics of an increase of 2.9 cm in occipital process distance and 3.1 cm in cervical depth distance respectively. The results of 3-D modeling further revealed various visual shape differences. In the process of establishing an intelligent classification model for young people's heads, necks and shoulders using machine learning, the SMOTE algorithm was employed to generate minority class samples through interpolation, in order to balance the category distribution of the training set. The linear model (SVM/LR) was significantly superior to the tree model (RF). All models tended to choose stronger regularization (C=10), and SVM selected the linear kernel. the study analysis indicated obvious linear relationships among the features. The performance of all model test sets was lower than that of cross-validation, and the random forest decreased most significantly (13.85%), indicating that the tree model was severely overfitted. The prediction results of the base models showed a high degree of similarity, and the voting ensemble did not surpass the best base model (SVM/LR). The overall accuracy rate of the intelligent classification model based on ensemble learning was found to be 92%, reaching an excellent level. The recognition effects of most categories were excellent, with 100% for the low-pillow flat-neck type, 94% for the normal type, and 92% for the deep-neck type.

Conclusion Based on 3-D human body scan data and cluster analysis, this study classified the young group into four categories, aiming to improve the deficiencies of existing head, neck and shoulder research in the fields of transgender and intelligent classification. By integrating 3-D scanning, cluster analysis, modeling techniques and machine learning, precise basis for the development of personalized engineering products (such as sleep pillows) was formulated. In the future, it is necessary to further verify the applicability of the classification system among a wider range of people (such as different age and occupational groups), and explore the influence of dynamic postures on morphology.

Key words: youth, three-dimensional human body scanning, head-neck-shoulders, three-dimensional modeling, intelligent classification model