Style-Aware Hierarchical Recognition for Accessible Museum Guidance Systems
Journal of Biomedical Research & Environmental Sciences · 2026-07 · 7 (7) · 1-13
DOI: 10.37871/jbres2319 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Biomedical Research & Environmental Sciences, 2026-07. Verfasst von Youngbok Cho.
Abstract (Originalsprache)
Background
Museum artwork recognition presents distinct challenges: extreme data scarcity (typically one training image per title), museum-specific imaging conditions, and high intra-class visual variability that requires sensitivity to artistic style rather than semantic content alone.
Methods
We present a Style-Aware Hierarchical Recognition Framework that addresses these challenges through three coordinated components. First, a domain-specific augmentation pipeline generates ten synthetic variants per original image, simulating spot lighting, lens distortion, motion blur, and partial occlusion to enrich training diversity. Second, the Style-Aware Feature Fusion (SAFF) module is mounted atop an Inception ResNetV2 backbone; it extracts a 256-dimensional global style embedding via average pooling, generates channel-wise attention weights through a sigmoid-activated dense layer, and reweights the multi-scale feature map in a residual manner to emphasize stylistically discriminative cues such as brushstroke texture, color layering, and compositional rhythm. Third, a two-stage hierarchical classifier first assigns each artwork to one of four departments, then applies a department-specific title-level classifier, reducing the effective classification space from 261 to 42–86 classes per model.
Results
On the Metropolitan Museum of Art benchmark, the framework achieves 86% department-level accuracy (F1 = 0.86) and 84% weighted-average Top-1 accuracy across 261 title classes (F1 = 0.81). The end-to-end hierarchical pipeline yields 75% final Top-1 accuracy, substantially outperforming flat baselines including InceptionResNetV2 alone (56%), InceptionV3 (45%), and VGG16 (37%). Ablation analysis confirms that the style embedding is the dominant performance factor (?9% when removed), and SAFF outperforms established attention mechanisms (SE-Net, CBAM, and transformer self-attention) under identical training conditions.
Conclusion
Combining domain-aware augmentation, style-sensitive feature fusion, and hierarchical decomposition provides a robust and scalable foundation for museum artwork recognition under extreme data constraints, with direct applicability to accessible museum guidance systems.
Zitieren
APA
Cho, Y. (2026). Style-Aware Hierarchical Recognition for Accessible Museum Guidance Systems. Journal of Biomedical Research & Environmental Sciences, 7(7), 1-13. https://doi.org/10.37871/jbres2319
Vancouver
Cho Y. Style-Aware Hierarchical Recognition for Accessible Museum Guidance Systems. Journal of Biomedical Research & Environmental Sciences. 2026;7(7):1-13. doi:10.37871/jbres2319
BibTeX
@article{Cho2026,
author = {Cho, Youngbok},
title = {Style-Aware Hierarchical Recognition for Accessible Museum Guidance Systems},
journal = {Journal of Biomedical Research & Environmental Sciences},
year = {2026},
volume = {7},
number = {7},
pages = {1--13},
doi = {10.37871/jbres2319}
}
EUVIMED