Clothing digital design evaluation system based on deep learning and multi-source information

Iranian Journal of Science and Technology, Transactions of Electrical Engineering · Springer Nature · 2026-11-19 · DOI: 10.1007/s44163-026-02531-2

Аннотация

Background The fashion industry is rapidly transforming with digital technologies and AI. Traditional design evaluation is subjective, inconsistent, and time-consuming. Key challenges include systematically assessing diverse clothing features and modeling complex relationships among visual aesthetics, 3D garment structures, fabric properties, and consumer preferences to enable accurate, data-driven design evaluation. Aim To address these challenges, this research presents a Clothing Digital Design Evaluation System that leverages deep learning and multi-source information to analyze and assess fashion designs. Method A multimodal feature representation is constructed with 10,335 fashion designs, integrating 2D design sketches, 3D virtual garment simulations, textile property data, and trend-related information. Pre-processing techniques, including image resizing and min-max normalization, are used to standardize data. The Convolutional Neural Network (CNN) is used to automatically extract the visual features. A deep learning framework includes Emperor Penguins Colony mutated Gated Recurrent with Conditional Variation Autoencoder (EPC-GR-ConVAE). The ConVAE is used for fit and drape analysis, and GR models for textual trends and EPC optimization for mapping the extracted features into a design evaluation Space. This enables a comprehensive evaluation of clothing aesthetics, sustainability, and predicted market acceptance. Result The suggested EPC-GR-ConVAE has been implemented in Python, and the experiments have proved that the solution outperforms the Random Forest (RF) model and yields superior outcomes, providing an accuracy of 95 to 97%, to optimize clothing designs. The presented results validate the model’s performance within the defined experimental setup; however, they do not constitute evidence of full real-world deployment or large-scale generalization. Conclusion Experimental evaluation demonstrates improved performance compared with baseline methods under defined classification settings. The results indicate that the model captures relationships between visual features and design attributes within the experimental scope. However, the applicability of the framework is currently limited to controlled conditions and depends on dataset quality.

Читать у издателя →

Похожие статьи