Deep Learning-based Severity Classification of Concrete Cracks using YOLOv8 for Structural Health Analysis

Abstract

Crack detection and severity classification are essential tasks in structural health monitoring, especially for critical civil infrastructure such as roads and bridges. Traditional methods rely heavily on manual inspection, which is time-consuming, costly, and prone to human error. This study introduces a computer vision approach using the YOLOv8n model to automatically classify concrete surface conditions into seven categories, ranging from "No Crack" to "Very Large Crack". After augmenting and preprocessing the dataset, the model was trained over 10 epochs and achieved an accuracy of 97.1% and accuracy of 99.9% on the validation set, demonstrating strong classification performance on 53 samples. With a dataset containing over 11,501 images across six categories, the model displayed strong generalization and fast inference speeds reached 0.4ms per image. These results validate the YOLOv8 classifier’s capability for rapid and accurate infrastructure assessment and pave the way for scalable deployment on drones and mobile devices in real-time field scenarios.