Image Processing Based Deep Learning Approach for Classifying Hybrid Filler Ratio in Glass Fiber Reinforced Polymer

In this study, an image processing-based deep learning approach is developed to classify the filler ratios in glass fiber reinforced polymer composite with hybrid MgO-CuO nanoparticles in a non-contact and non-destructive manner, based solely on surface color and texture information. The originality of the study lies in its reliance on the direct learning of hybrid nanoparticle dopant ratios from optical surface properties, unlike image-based methods in the literature that mainly focus on damage detection, phase separation, or mechanical property estimation. In this context, eight different composite classes with different MgO and CuO weight ratios were produced and the samples were imaged at high resolution under homogeneous LED illumination at a fixed camera-sample distance. The obtained images were evaluated in a multi-class classification problem using the transfer learning-based EfficientNet-B0 architecture without any data enhancement. Model performance was analyzed with accuracy, sensitivity, recall, and F1-score metrics. Test results show that the proposed model achieved 97% overall accuracy and a macro-mean F1-score of 0.97; The study demonstrated that single-dopant and high-contrast hybrid systems were classified with high accuracy rate. Limited class overlap was observed in some hybrid classes with low CuO content, which is attributed to the optical similarity of the dopants. The findings reveal that deep learning approaches based on surface images offer a powerful, low-cost, and industrially viable alternative for dope ratio verification and rapid quality control applications in composite manufacturing processes.