DeepMarbleVision: A Texture-Aware Ensemble Deep Learning Model with Energy-Layer-Based Feature Fusion for Marble Classification
COMPUTERS, MATERIALS AND CONTINUA, cilt.2026, sa.1, ss.1-20, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 2026 Sayı: 1
- Basım Tarihi: 2026
- Doi Numarası: 10.32604/cmc.2026.085198
- Dergi Adı: COMPUTERS, MATERIALS AND CONTINUA
- Derginin Tarandığı İndeksler: Scopus, Science Citation Index Expanded (SCI-EXPANDED), Compendex, INSPEC, zbMATH
- Sayfa Sayıları: ss.1-20
- Sivas Cumhuriyet Üniversitesi Adresli: Evet
Özet
Marble classification has traditionally relied on human visual inspection, where operators assess color, texture, and pattern alignment to determine quality. However, this manual process is subjective, inconsistent, and inefficient for large-scale industrial applications. To address these limitations, this study proposes DeepMarbleVision, a texture-aware ensemble deep learning framework with energy-layer-based feature fusion for marble quality classification. A real-world dataset was created using the MarbleVision system, including three marble quality classes acquired from an industrial marble classification environment. The proposed approach integrates energy-layer-based feature fusion into TCNN variants of AlexNet, ResNet, and DenseNet, which were initialized through texture-oriented pre-training and fine-tuned for marble quality classification. To further improve classification robustness, an ensemble learning strategy was applied by averaging the class-probability outputs of individual CNN and TCNN models. The ensemble model combining baseline CNN and energy-enhanced TCNN architectures achieved 99.5% classification accuracy, outperforming the evaluated standalone TCNN models: AlexNet-TCNN, 89.25%; DenseNet-TCNN, 95.16%; and ResNet-TCNN, 92.83%. These findings indicate that energy-layer-enhanced ensemble deep learning models can improve texture-based marble quality classification compared with the evaluated standalone CNN and TCNN models. The proposed model is intended for future integration into the MarbleVision automated marble classification pipeline and provides an adaptable framework for high-precision aesthetic surface inspection in related industrial applications.