Non-Destructive Assessment of Concrete Air Permeability Based on Texture Analysis and SVM Classification

This paper explores the use of machine learning as an efficient tool for classifying the air permeability of concrete. The classification was based on the air permeability coefficient kT, previously determined using the Torrent tester. According to this coefficient, the tested concrete mixtures were classified as having "Poor" and "Very Poor" cover layer quality. For feature extraction, three approaches were investigated: Histogram of Oriented Gradients (HOG), Gabor filters, and Gray-Level Co-occurrence Matrix (GLCM) descriptors. A nonlinear Support Vector Machine (SVM) classifier was used to perform the classification task based on the obtained features. The SVM classifier was trained on a dataset derived from 4,333 original images of concrete surfaces. To improve the robustness of the training process, data augmentation was applied, resulting in 13,464 training samples, while the model was evaluated on a separate test set of 867 images. The performance of the proposed model was assessed using commonly adopted evaluation metrics, including accuracy, precision, recall, F1-score, and ROC-AUC. Among the tested feature descriptors, the Gabor-based representation achieved the best overall performance. The experimental results on the test set yielded an accuracy of 0.9527, precision of 0.9481, recall of 0.9826, an F1-score of 0.965, and a ROC-AUC of 0.9874. The results are sufficiently strong and promising, establishing a solid basis for the development of more efficient and effective methods in the field of concrete assessment.