A Novel Hybrid Feature Optimization Framework for Road Surface Identification with Vision Transformer and EGWO-SVM

Accurate identification of road surface conditions is essential for autonomous driving, Intelligent Transportation Systems (ITS), and driver-assist technologies. This paper proposes an integrated hybrid framework that combines multi-resolution texture decomposition, attention-based deep learning, handcrafted statistical descriptors, and evolutionary feature optimization. The approach begins by applying Stationary Wavelet Transform (SWT) to extract stable texture characteristics from road images. A Vision Transformer enhanced with Squeeze-and-Excitation (ViT-SE) blocks captures global spatial relationships, whereas Gray-Level Co-Occurrence Matrix (GLCM) descriptors contribute fine-grained statistical features. These complementary features are filtered using Mutual Information (MI) and optimized using Enhanced Grey Wolf Optimization (EGWO), enabling compact and discriminative feature selection. The optimized feature subset is classified using a Support Vector Machine (SVM). Experiments conducted on a multi-class road surface dataset demonstrate that the proposed SWT–ViT-SE–GLCM–MI–EGWO–SVM pipeline achieves 99.52% accuracy, outperforming state-of-the-art Convolutional Neural Network (CNN), Transformer, and hybrid models. The results confirm the synergy of multi-domain features and metaheuristic optimization in real-time road condition assessment applications.