Enhancing Pavement Distress Classification via Supervised Contrastive Learning Denetimli Karşitlikli Ö?grenme Yoluyla Yol Kaplama Bozulma Siniflandirmasinin ?Iyileştirilmesi
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636991
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: Computer Vision, Deep Learning, Pavement Distress, Supervised Contrastive Learning
- Ankara Yıldırım Beyazıt Üniversitesi Adresli: Evet
Özet
The performance of deep learning models in pavement distress detection is often limited by data scarcity and interclass visual similarities. This study proposes a hybrid training strategy combining Cross-Entropy and Supervised Contrastive Learning to enhance discriminative feature learning. Experiments conducted on the PaveDistress dataset using ResNet50 and EfficientNetB0 architectures demonstrate that the proposed method outperforms traditional fine-tuning approach. The EfficientNetB0 model achieved a test accuracy of 92.74% (a 1.63% improvement), and t-SNE analyses confirmed significantly improved class separability and clustering quality.