Concrete structures are a vital component of urban infrastructure, requiring regular maintenance to ensure public safety and structural integrity. A crucial element of this maintenance is the identification of surface cracks, which have traditionally relied on manual inspection methods that were frequently work-intensive, subjective, and sometimes dangerous. This work presents a hybrid methodology that integrates deep feature extraction with machine learning classification for identifying structural deterioration in concrete components. A publicly accessible dataset comprising photos of both cracked and uncracked concrete surfaces was used. Deep features were extracted using VGG16, a convolutional neural network widely recognized for its success in visual pattern recognition. Several machine learning algorithms were used for classification of these features, including Artificial Neural Network, Decision Tree, Random Forest, Support Vector Machines and k-Nearest Neighbors. The experimental results indicate that, the highest accuracy was achieved by SVM (99.883%), followed closely by ANN (99.873%), k-NN (99.598%), and DT (99.580%), while RF performed the lowest (98.050%). Although not limited to seismic applications, the proposed method has the potential to be integrated into post-earthquake structural assessment workflows as part of structural health monitoring systems. Using deep learning and machine learning methodologies to detect damage in concrete infrastructure may enhance efficiency and precision, enhancing urban resilience and risk mitigation.
Niğmet Köklü (Sat,) studied this question.