Road crack detection using deep neural networks developed via cooperative coevolution with backpropagation

Cracks are superficial damages that occur from the substrate to the pavement surface due to the effect of traffic loads. Detecting a crack damage before it grows and performing the necessary maintenance contributes positively to both the road comfort and the expenses to be made for maintenance. In this study, it is aimed to detect the cracks on the road in real time and with high accuracy. In this context, Deep Neural Networks Developed via Cooperative Coevolution with Backpropagation and image processing and image processing methods were used together. In the study, a new data set was obtained by using EdmCrack600, AsphaltCrack, CFD and CrackSegmentation datasets containing cracked visual data in various numbers and resolutions, and Deep Neural Networks-based learning was performed on this dataset. Thanks to image processing techniques, objects without cracks were removed from the image in which cracks were detected, and a black-and-white picture showing the rough location of the crack was obtained. Finally, area-based crack detection was performed by using the parameters of the network structure that performed the best learning on the roughly positioned crack. The accuracy of the model was evaluated with Precision, Sensitivity and F1-Score criteria using CFD dataset. As a result of the evaluation,it has been observed that the proposed method can detect cracks on 48 images per second, while it can reach 92.74% Precision, 88.92% Recall and 89.61% F1-Score success rates.