Hybrid quantum‐classical convolutional models for image‐based infrastructure inspection and assessment

• Published in: Computer-aided civil and infrastructure engineering Vol. 40; no. 24; pp. 3894 - 3910
• Main Authors: Niu, Yujun, Wang, Chaofeng
• Format: Journal Article
• Language: English
• Published: Hoboken Wiley Subscription Services, Inc 01.10.2025
• Subjects: Ablation; Coding; Configuration management; Feasibility studies; Image classification; Infrastructure; Inspection; Machine learning; Noise measurement; Parameters; Quantum computing
• ISSN: 1093-9687; 1467-8667
• DOI: 10.1111/mice.70038

Infrastructure inspection and assessment such as pavement distress classification remains a critical challenge for infrastructure maintenance. Traditional manual inspections are labor‐intensive, costly, and subject to subjective bias, while classical machine learning models demand extensive computational resources for large‐scale evaluations. These challenges have driven growing interest in quantum computing, which theoretically offers increased computational efficiency and the potential to process high‐dimensional data with fewer parameters compared to classical approaches. This study investigates the feasibility and performance of Hybrid Quantum‐Classical Convolutional Models (HQCCMs) specifically designed for infrastructure image classification. Two distinct HQCCMs are proposed: the Parallel Angle‐Encoded Hybrid Quantum‐Classical Convolutional Model (PA‐HQCCM) and the Single Amplitude‐Encoded Hybrid Quantum‐Classical Convolutional Model (SA‐HQCCM). The core novelty of these hybrid architectures lies in the effective integration of quantum convolutional layers, classical convolutional operations, and parameterized quantum circuits (PQCs) into a unified framework, enabling the models to leverage the inherent compactness and representational efficiency of quantum feature extraction. Experimental results reveal that HQCCMs achieve a test accuracy improvement of approximately 3.0% beyond comparable classical CNN baselines, while significantly reducing the number of trainable parameters. Comprehensive ablation studies quantify the distinct advantages delivered by the quantum components, while generalization capabilities are assessed by unseen images. The study further examines the influence of measurement strategies, quantum noise, and circuit configurations on model performance, offering valuable insights into optimizing HQCCMs. Overall, these findings underscore the transformative potential of quantum‐enhanced architectures in resource‐constrained environments, offering a scalable, image‐driven tool for rapid pavement condition screening.