Eggshell Crack Classification Using a Hybrid Texture-Color Descriptor with Machine Learning Methods

This study presents a lightweight and efficient method for classifying eggshell fractures as an alternative to deep learning-based approaches. The dataset was compiled from public and field sources, resulting in a balanced collection of 3,700 RGB images (1,850 fractured and 1,850 normal). The images were preprocessed by cropping, resizing to 256×256, and enhancement using CLAHE and GrabCut segmentation. Color (HSV), edge (Canny), and texture (GLCM and Gabor) features were combined into a hybrid descriptor and classified using Support Vector Machine (SVM), Random Forest (RF), XGBoost, and a stacking ensemble under 5-fold cross-validation. RF demonstrated the strongest robustness to nonlinear and texture-rich attributes, achieving the highest accuracy of 81.2%. The results confirm that the proposed hybrid descriptor offers a computationally lightweight, interpretable, and CPU-friendly alternative to CNN-based systems. Feature importance and PCA analysis further reveal how color and texture descriptors jointly contribute to decision-making, supporting its practicality for real-time industrial deployment.