Researchers have developed a hybrid artificial intelligence model that combines Convolutional Neural Networks and Vision Transformers to detect and classify pavement distress with high precision. The study, published in Neural Computing and Applications, presents an enhanced technique for road maintenance that aims to address the limitations of traditional visual inspection methods. By integrating the strengths of both deep learning architectures, the team achieved a precision of 99.02% and an accuracy of 99.89% in their experiments. This approach represents a significant improvement over using either CNNs or Vision Transformers individually for this specific task.
Pavement distress remains a major concern for road maintenance worldwide, posing risks to road users and generating high maintenance costs. Despite advances in material durability, issues such as heavy traffic, environmental and climatic conditions, and poor workmanship continue to degrade road surfaces. The traditional method of assessing these conditions relies on visual inspection, which is labor-intensive and prone to human error. These factors often result in delays in both the assessment of road conditions and the subsequent repair interventions.
Existing machine learning approaches for detecting and classifying pavement distress have often faced performance limitations. Additionally, these models are highly susceptible to overfitting. To address these challenges, the research team proposed a hybrid model that leverages the characteristics of local and global hierarchical images. This ensemble method utilizes the integration of CNN architectures with Vision Transformers to detect various types of pavement degradations more effectively.
The database for this study was collected through a planned inspection conducted in partnership with the Moroccan Road and Bridge Research Establishment, also known as CNER, under the Ministry of Equipment. A variety of pavement images were obtained using an intelligent inspection vehicle equipped with a multifunctional road assessment imaging system. This system incorporated a rear-view camera.
The experimental results demonstrated that the integrated methodology outperformed the individual CNN and ViT models. The ensemble model achieved a precision of 99.02% and an accuracy of 99.89% in detecting and classifying pavement distress. These metrics indicate that the hybrid approach provides a highly accurate tool for automating pavement inspection programs.
The study highlights the potential of this model to help road managers optimize inspection and maintenance interventions. By providing accurate and efficient detection of pavement distress, the technology could contribute to ensuring optimized model performance in infrastructure engineering. The findings are particularly relevant for pavement inspection companies and organizations responsible for the automatic evaluation of road surface conditions.
The use of Vision Transformers alongside Convolutional Neural Networks allows the model to capture both local details and global context within the images.
Subject of Research: Development of a hybrid CNN-ViT model for pavement distress detection
Article Title: Improved detection and classification of pavement distress adopting convolutional neural networks—vision transformer architecture
Article References: Aouni, Y., Ogri, O. E., Alami, S. E. M. E., EL-Mekkaoui, J., Hjouji, A., & Qachar, M. (2026). Improved detection and classification of pavement distress adopting convolutional neural networks—vision transformer architecture. Neural Computing and Applications, 38(21), Article 879. https://doi.org/10.1007/s00521-026-12601-2
Image Credits: AI Generated
DOI: 10.1007/s00521-026-12601-2
Keywords: pavement distress, convolutional neural networks, vision transformers, road maintenance, machine learning, infrastructure engineering, automated inspection, deep learning, Improved, detection, classification, pavement
News Source: Denise Maddox. (November 9, 2026). Hybrid AI Model Detects Pavement Distress with High Accuracy. Scienmag.



