Determination of Pavement Distress Severity Using Machine Learning: Systematic Review

Artificial intelligence techniques have advanced significantly in recent years and can be applied across various fields of knowledge. In civil engineering, such techniques as machine learning and computer vision have proven useful in pavement inspection and evaluation—processes that are traditionally carried out manually in the field, demanding considerable time and effort from inspectors. To address the limitations of manual inspections, numerous studies have aimed to automate this process by employing algorithms capable of recognizing and classifying surface distress. However, few studies address these algorithms’ ability to determine the severity level of pavement distress, a crucial piece of information for planning maintenance and rehabilitation activities. Therefore, the aim of this study is to conduct a systematic literature review, using the systematic search flow method, of research that employed machine learning techniques to automatically or semi-automatically determine the severity levels of pavement distresses. A search conducted in two databases (Scopus and Web of Science) yielded a total of 283 articles. After applying a defined filtering process, 28 studies were identified as meeting the established scope. A review of these articles revealed that machine learning is an effective method for distress recognition and severity classification. Nonetheless, owing to the limited exploration of this topic in the current literature, further investigation and detailed analysis of specific elements are necessary in future studies.