Faster R-CNN structure for computer vision- based road pavement distress detection

Aim The aim of this study is to detect cracks on asphalt fully automatically using Faster R-CNN structure. Design & Methodology For this study, photographs of various asphalt cracks are used for the dataset. The size of the dataset is one of the factors that increase performance in deep learning methods. For this reason, the dataset has been expanded using the perspective transform. Then, Faster R-CNN structure was created using Python programming language. Performance analyzes were made with the data obtained as a result of the test. Originality Faster R-CNN structure was used to detect asphalt cracks in the study. Findings The average precision for all tags is greater than 0.90, of which Non-crack gets the highest score of 94% and the Longitudinal Crack gets the lowest score of 92%. Mean precision score (mAP) of all tag is 93.2%. Finally, the Kappa value is 0.915. Conclusion As a result of the study, it has been determined that the Faster R-CNN structure, which is one of the deep learning techniques, can be used in the detection of cracks on asphalt. Declaration of Ethical Standards The authors of this article declare that the materials and methods used in this study do not require ethical committee permission and/or legal-special permission.