Akademik Veri Yönetim Sistemi
Optics and Lasers in Engineering, cilt.198, 2026 (SCI-Expanded, Scopus)
Material-specific spectral reflectance provides a reliable basis for identification and classification. Based on this principle, we offer a low-cost, three-wavelength, distance-scanning fiber optic system that is ideal for material identification, surface defect inspection, and quality control in confined or difficult-to-access industrial settings. In this study, we developed a compact, cost-effective, optical fiber non-contact object classification (OF-NOC) using three distinct wavelengths. Reflectance data collected from ten objects is used to train and test various machine and deep learning classifiers, including a narrow-layered neural network (NL-NN), a bilayered NN (BL-NN), a trilayered NN (TL-NN), a weighted K-nearest neighbors (WKNN), a support vector machine (SVM), a convolutional neural network (CNN), a gated recurrent unit (GRU), and a long short-term memory (LSTM). The ten objects were restructured into four material-based classes to evaluate generalization performance. For ten objects, the GRU model achieved the highest average accuracy (0.939), followed closely by the TL-NN (0.919) and cubic SVM (0.913). The proposed OF-NOC system demonstrates strong classification performance and has advantages such as portability, scalability, and hardware simplicity. Thanks to its compact structure, low-cost design, and proven performance, the system provides a scalable solution for industrial quality control, robotic sensing, and precise object classification applications.