Mermer yüzeylerinin otomatik sınıflandırılmasında transformer tabanlı mimarilerin karşılaştırmalı analizi

Abstract

Bu çalışmada, mermer yüzeylerinin renk ve doku özelliklerine dayalı olarak tür sınıflandırmasının yapılması amaçlanmış; bu doğrultuda görüntü işleme görevlerinde giderek daha güçlü bir alternatif haline gelen Transformer mimarilerinin performansı incelenmiştir. Araştırma kapsamında, 25 farklı mermer türüne ait toplam 5600 görüntüden oluşan bir veri seti kullanılarak Vision Transformer ve Swin Transformer modelleri değerlendirilmiştir. Görüntüler ön işleme adımlarından geçirilerek 224x224 piksel boyutuna getirilmiş; ardından yansıtma, döndürme, parlaklık ve kontrast değişiklikleri gibi veri artırımı teknikleri uygulanarak veri seti 18.275 görüntüye genişletilmiştir. Elde edilen deneysel bulgulara göre, Swin Transformer modeli orijinal veri setinde %97.91 doğruluk oranı gösterirken, veri artırımı uygulandıktan sonra bu oranın %98.74 seviyesine ulaştığı gözlemlenmiştir. Ayrıca, F1 metriği açısından da Swin Transformer'ın oldukça yüksek bir performans sergilediği görülmüştür. Bu bulgular ışığında, Transformer tabanlı modellerin mermer yüzeylerindeki karmaşık doku ve renk dağılımlarını öğrenmede etkili bir araç olabileceği ve geleneksel evrişimsel sinir ağı yöntemlerine esnek bir alternatif sunabileceği değerlendirilmektedir.

Keywords

• Vision transformer
• Swin transformer
• Görüntü işleme
• Mermer sınıflandırma

A comparative analysis of transformer-based architectures for the automatic classification of marble surfaces

Abstract

The aim of this study was to classify marble types based on their color and texture characteristics; to this end, the performance of Transformer architectures—which are increasingly becoming a powerful alternative in image processing tasks—was examined. As part of the research, Vision Transformer and Swin Transformer models were evaluated using a dataset consisting of a total of 5,600 images representing 25 different marble types. The images were preprocessed to a size of 224x224 pixels; subsequently, the dataset was expanded to 18,275 images by applying data augmentation techniques such as reflection, rotation, and adjustments to brightness and contrast. According to the experimental findings, the Swin Transformer model achieved an accuracy rate of 97.91% on the original dataset, and this rate was observed to reach 98.74% after data augmentation was applied. Additionally, the Swin Transformer was found to demonstrate quite high performance in terms of the F1 metric. In light of these findings, it is assessed that Transformer-based models could serve as an effective tool for learning the complex textures and color distributions on marble surfaces and offer a flexible alternative to traditional convolutional neural network methods.

Keywords

• Vision transformer
• Swin transformer
• Image processing
• Marble classification

References

1. Demir, B. G., & Güngör, N. (2013). Mermer Madenciliği ve Çevre. İstanbul Aydın Üniversitesi Dergisi, 5(20), 7-14.
2. Yalçın, S., & Uyanık, T. (2001). Dünya Mermer Ticaretinde Türkiye'nin Yeri.
3. Çetin, T. (2003). Türkiye mermer potansiyeli, üretimi ve ihracatı. Gazi Üniversitesi Gazi Eğitim Fakültesi Dergisi, 23(3).
4. Bianconi, F., González, E., Fernández, A., ve Saetta, S. A. (2012). Automatic classification of granite tiles through colour and texture features. Expert Systems with Applications, 39(12), 11212–11218.
5. Canayaz, M., ve Uludağ, F. (2020). Marble classification using deep neural networks. European Journal of Technique , 10(1), 52–63.
6. Gürkan, Ç., ve Palandöken, M. (2021). SMARfacTory-Net: Mermerin sınıflandırılması için bilgisayarlı görü, QR kod ve Android tabanlı teknolojilerle desteklenen sistem tasarımının geliştirilmesi. Computer Science (Special Issue), 347–358.
7. Sharma, N., Jain, V., & Mishra, A. (2018). An analysis of convolutional neural networks for image classification. Procedia computer science, 132, 377-384.
8. Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., ... ve Polosukhin, I. (2017). Attention is all you need. Advances in Neural Information Processing Systems, 30, 5998–6008.

9. Song, C. H., Han, H. J., & Avrithis, Y. (2022). All the attention you need: Global-local, spatial-channel attention for image retrieval. In Proceedings of the IEEE/CVF winter conference on applications of computer vision (pp. 2754-2763).
10. Elmas, B. (2022). Classification varieties of marble and granite by convolutional neural networks with transfer learning method. Journal of the Faculty of Engineering and Architecture of Gazi University, 37(2), 985–1001.
11. Yavuz, M., ve Türkoğlu, İ. (2025). Determination of the quality classes of Elazig cherry marble with image processing. Ain Shams Engineering Journal, 16(8), 103455.
12. Ouzounis, A. G., Taxopoulos, G., Papakostas, G. A., Sarafis, I. T., Stamkos, A., & Solakis, G. (2021, October). Marble quality assessment with deep learning regression. In 2021 Fifth International Conference On Intelligent Computing in Data Sciences (pp. 1-5). IEEE.
13. Gayathri, S., Gowthami, J., Jose, J. J., & Kowsalya, D. (2026, February). Quality Assessment of Marble using Deep Learning Techniques. In 2026 International Conference on Electronics and Renewable Systems (pp. 1330-1335). IEEE.
14. Gonzalez, R. C. (2009). Digital image processing. Pearson Education India.
15. Chen, T. (1998). The past, present, and future of image and multidimensional signal processing. IEEE Signal Processing Magazine, 15(2), 21–58.
16. Karaca, A., ve Aydın, Ö. (2023). Transformatör mimarisi tabanlı derin öğrenme yöntemi ile Türkçe haber metinlerine başlık üretme. Gazi Üniversitesi Mühendislik Mimarlık Fakültesi Dergisi, 39(1), 485–496.
17. Han, K., Wang, Y., Chen, H., Chen, X., Guo, J., Liu, Z., ... ve Tao, D. (2022). A survey on vision transformer. IEEE Transactions on Pattern Analysis and Machine Intelligence, 45(1), 87–110.
18. Dosovitskiy, A. (2020). An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929.
19. Liu, Z., Lin, Y., Cao, Y., Hu, H., Wei, Y., Zhang, Z., ... ve Guo, B. (2021). Swin transformer: Hierarchical vision transformer using shifted windows. Proceedings of the IEEE/CVF International Conference on Computer Vision (ss. 10012–10022).
20. Kanca, E., Ayas, S., Kablan, E. B., ve Ekinci, M. (2023, Temmuz). Performance comparison of vision transformer-based models in medical image classification. 31st Signal Processing and Communications Applications Conference (ss. 1–4). IEEE.
21. Abadi, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean, J., ... & Zheng, X. (2016). {TensorFlow}: a system for {Large-Scale} machine learning. In 12th USENIX symposium on operating systems design and implementation (pp. 265-283).
22. Prechelt, L. (2002). Early stopping—but when?. In G. Orr & K. Müller (Eds.), Neural Networks: Tricks of the Trade (ss. 55–69). Berlin, Heidelberg: Springer.
23. Hossin, M., ve Sulaiman, M. N. (2015). A review on evaluation metrics for data classification evaluations. International Journal of Data Mining & Knowledge Management Process, 5(2), 1–9.