Oppositional White Shark Optimizer for Classifying Road Cracks in Smart City Infrastructure
Abstract
Smart city infrastructure development uses high-tech road surface quality monitoring systems to optimize maintenance schedules, save operating costs, and improve safety. Support vector machines (SVMs) and convolutional neural networks (CNNs), the two major methods for detecting road cracks, typically fail due to their complexity in handling surface textures and a lack of generalizability. To overcome these restrictions, this study presents the Oppositional Shark-based Crack Classification Network (OppShark-CrackNet), a hybrid architecture that combines Vision Transformers (H-ViT) with Graph Convolutional Networks (GCNs), optimized by an OWSO. The GCN employs a relational topology between features, whereas the H-ViT module identifies long-range connections and structural abnormalities to capture high-level fracture characteristics and achieve robust classification. Oppositional White Shark Optimizer (OWSO) optimizes hyperparameters and dynamically selects the best discriminative feature embeddings. The proposed method used the public Kaggle Potholes or Cracks on Road Image Dataset to assess the suggested approach. The proposed OppShark-CrackNet outperforms metaheuristic-based classifiers and state-of-the-art models, such as DenseNet and Swin Transformer, in F1-score and accuracy by 7-10%. Our technique is well-suited for real-time edge-based smart city monitoring systems due to its noise robustness and faster convergence. Automatic road infrastructure management and next-generation predictive maintenance are possible using OppShark-CrackNet’s deep transformer representations and graph-based reasoning.