Umeå University's logo

The preservation of structural integrity and durability is essential for the long-term viability of civil infrastructure projects. Addressing concrete defects promptly is crucial to achieving this objective. In this research, the research proposes a novel method for concrete defect analysis, harnessing the potential of image processing and deep learning techniques. It employs a fusion-based deep convolutional neural network (CNN), amalgamating the features of Inception V3, VGG16, and AlexNet architectures to identify and classify six distinct concrete defect characteristics, namely Cracks, Crazing, Efflorescence, Pop-out, Scaling, and Surface Cracks. Through rigorous training and validation, we thoroughly assess the performance of the proposed fusion-based CNN model. The testing phase reveals precision rates, with Crazing showing the lowest precision at 95%, and Cracks/Pop-outs achieving 98%, while other defect classifications exhibit exceptional 100% precision rates. The overall efficacy of our proposed model is evaluated using accuracy and F1-score metrics. Our findings demonstrate an impressive overall accuracy of 98.31% and an F1-score of 0.98, affirming the robustness and reliability of our approach. The outcomes of this study signify a significant advancement toward accurate and automated detection and classification of concrete defects.