Work

Multi-scale residual network-driven regional cultural color feature extraction and intelligent cultivation pathway for rural art design talent

Year: 2026

Type: article

Abstract: The preservation of regional cultural color heritage faces mounting challenges from urbanization and aesthetic homogenization, while rural art design education lacks effective integration of intelligent technologies. This paper proposes a unified framework addressing both computational color analysis and talent cultivation for rural cultural revitalization. On the technical side, we design a multi-scale residual network (MSResNet) that employs parallel dilated convolutional branches with channel attention-weighted fusion to capture hierarchical chromatic features from regional cultural artifacts. A cross-modal semantic association mechanism further embeds extracted color features into a culturally annotated latent space, enabling context-aware interpretation beyond purely statistical description. On the educational side, we construct a three-stage intelligent cultivation pathway—comprising cognitive tutoring, database-driven practice, and personalized competency assessment—that operationalizes the model’s outputs within a pedagogically grounded framework. Experiments on the purpose-built RCC-6 K dataset demonstrate that MSResNet achieves a macro F1 score of 0.912 and a cultural association index of 0.831, outperforming representative baselines. A 16-week quasi-experimental pedagogical study with 86 vocational college students shows that the intelligent pathway yields a composite competency gain of 29.9%, significantly exceeding the 14.6% achieved under traditional instruction. These results confirm that structurally integrating multi-scale deep learning with semantic cultural grounding can advance both the computational characterization and educational transmission of regional color heritage.

Source: Scientific Reports

DOI: 10.1038/s41598-026-70860-4

Authors: Xiaoyan Lu, Zhengyin Gong

Institution: Quzhou College of Technology

Language: English

Topic: Advanced Neural Network Applications

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

SDG: Quality education

Open Access status: gold

Estimated APC paid: $2,690