View in EDS

Design and Functional Evaluation of Adaptive Clothing for Female Wheelchair Users Driven by Deep Learning

Saved in:
Bibliographic Details
Title: Design and Functional Evaluation of Adaptive Clothing for Female Wheelchair Users Driven by Deep Learning
Authors: Huizhi Jiang, Verly Veto Vermol, Hasma Bonti Ahmad, Si Chen
Source: IEEE Access, Vol 13, Pp 187724-187739 (2025)
Publisher Information: IEEE, 2025.
Publication Year: 2025
Collection: LCC:Electrical engineering. Electronics. Nuclear engineering
Subject Terms: Human-centered computing, transductive relational learning, curriculum optimization, adaptive systems, intelligent interfaces, Electrical engineering. Electronics. Nuclear engineering, TK1-9971
Description: This is particularly vital in the design of personalized wearable systems, where traditional approaches often rely on static, rule-based mechanisms that fail to adapt to complex, real-world constraints. Conventional techniques in adaptive design often fail to capture subtle relationships and contextual patterns relevant to real-world user needs, leading to models that generalize poorly across diverse user settings. To address this limitation, we present an integrated framework that leverages a Transductive Relational Encoder Network (TREN) and a Contextual Curriculum Optimization Protocol (CCOP). TREN encodes structural relationships among inputs via graph-based attention and gated message passing, enabling nuanced representation learning aligned with inter-instance semantics. Meanwhile, CCOP orchestrates a stage-wise training schedule that adapts dynamically to input complexity, using entropy-guided curriculum selection and temporal coherence alignment to regularize learning. Together, these innovations create a robust pipeline that improves discriminative performance while preserving contextual sensitivity, offering significant enhancements in functional adaptability and user-centric inference. This methodological synergy aligns with the journal’s interdisciplinary focus on intelligent interfaces, adaptive algorithms, and personalized computing systems. Experimental results show that our method achieves superior performance compared to existing baselines, with an average mAP improvement of 2.5–4.0 points across four benchmark datasets. Specifically, our approach reached an mAP of 80.76 on the CAESAR dataset and 88.53 on UP-3D, surpassing YOLOv8 and DETR by notable margins. These results validate the model’s enhanced adaptability, accuracy, and user-centric inference capabilities.
Document Type: article
File Description: electronic resource
Language: English
ISSN: 2169-3536
Relation: https://ieeexplore.ieee.org/document/11207680/; https://doaj.org/toc/2169-3536
DOI: 10.1109/ACCESS.2025.3623327
Access URL: https://doaj.org/article/8a2dbd072ebe4199837ef43a08a755e4
Accession Number: edsdoj.8a2dbd072ebe4199837ef43a08a755e4
Database: Directory of Open Access Journals
Description
Abstract:This is particularly vital in the design of personalized wearable systems, where traditional approaches often rely on static, rule-based mechanisms that fail to adapt to complex, real-world constraints. Conventional techniques in adaptive design often fail to capture subtle relationships and contextual patterns relevant to real-world user needs, leading to models that generalize poorly across diverse user settings. To address this limitation, we present an integrated framework that leverages a Transductive Relational Encoder Network (TREN) and a Contextual Curriculum Optimization Protocol (CCOP). TREN encodes structural relationships among inputs via graph-based attention and gated message passing, enabling nuanced representation learning aligned with inter-instance semantics. Meanwhile, CCOP orchestrates a stage-wise training schedule that adapts dynamically to input complexity, using entropy-guided curriculum selection and temporal coherence alignment to regularize learning. Together, these innovations create a robust pipeline that improves discriminative performance while preserving contextual sensitivity, offering significant enhancements in functional adaptability and user-centric inference. This methodological synergy aligns with the journal’s interdisciplinary focus on intelligent interfaces, adaptive algorithms, and personalized computing systems. Experimental results show that our method achieves superior performance compared to existing baselines, with an average mAP improvement of 2.5–4.0 points across four benchmark datasets. Specifically, our approach reached an mAP of 80.76 on the CAESAR dataset and 88.53 on UP-3D, surpassing YOLOv8 and DETR by notable margins. These results validate the model’s enhanced adaptability, accuracy, and user-centric inference capabilities.
ISSN:21693536
DOI:10.1109/ACCESS.2025.3623327