Road damage detection on Hatay province highways using machine learning methods

Currently, the maintenance and repair of highways generally rely on physical observations and personal initiatives. This slows down the correction of road damages. Road damages significantly affect the comfort and safety of drivers. The detection of road damages is important not only for transportation safety but also in terms of cost. Detecting road damages is crucial for enabling early intervention and repair. Therefore, many studies are being conducted to detect, identify, and classify road damages using machine learning techniques. In this study, the performance of the YOLOv8 algorithm for road damage detection was evaluated on datasets obtained from various geographies, including the Czechia -Turkey, India-Turkey, USA-Turkey, and Japan-Turkey. The findings reveal both the capabilities of the algorithm in damage detection and the challenges the model faces in distinguishing certain types of damages. Images of roads in the Hatay province were recorded to create the Turkey dataset, allowing for a detailed analysis of road damages. Labeling on the dataset was conducted using Microsoft's VoTT application. Machine learning algorithms are used for the identification of road damages. Evaluations and comparisons have been made among the developed models. The best result among the models was given by the Japan-Turkey model with a 0.55 mAP and 0.54 F1 score. Based on the values obtained from the models, it has been observed that the appearance of damages varies according to geography and road data. The necessity of giving more importance to training on less clear types of damages and local images has been observed.