Akademik Veri Yönetim Sistemi
Annals of Operations Research, 2026 (SCI-Expanded, Scopus)
Accurate detection of surface defects is crucial due to the complex background textures on material surfaces, significant intra-class defect variations, and the presence of subtle texture defect regions. This study introduces an innovative approach that employs the UNET model to extract image features from surface defect detection datasets. The extracted features are then selected using the African Vultures Optimization Algorithm (AVOA), Honey Badger Algorithm (HBA), and Whale Optimization Algorithm (WOA) metaheuristic algorithms, with classification carried out using Random Forest (RF) and Gradient Boosting (GB). The proposed approach is validated on the DAGM 2007 and KollektorSSD2 datasets. The experimental results show that the proposed approach is both efficient and effective, significantly outperforming supervised methods and achieving performance comparable to others. Specifically, it achieves accuracy rates of 82% with WOA + GB and 97% with AVOA + RF on the DAGM 2007 and KollektorSSD2 surface defect detection datasets, respectively. Additionally, an ablation study was also performed with feature extractors and classification algorithms, demonstrating that the proposed hybrid model achieves superior classification accuracy compared to state-of-the-art surface defect classification approaches. These findings highlight the robustness and adaptability of the proposed method, providing an efficient and dependable solution for surface defect classification in industrial settings.