Identify Faults in Road Structure Zones with Deep Learning

One of the most important causes that cause traffic accidents is the deterioration or failures in road structures, which occurs due to the low quality of its components, climate changes, and seismic zones, heavy cargo transport traffic, among others; therefore, the purpose of this research is the detection of faults in the pavement. The set of images used was classified into cracks and gaps, making a total of 420 images. The research was carried out taking into account the following stages: data collection, through a smartphone the images were captured; data preprocessing, which allowed the best images to be selected and redimensioned; model of the network architecture, allowed the selection and improvement of the algorithm; training and verification of the algorithm, to create an optimal model in the detection of cracks and gaps, as a last stage, the deployment, where the system for the test of the model was developed; for which a Convolutional Neural Network (CNN) was used as the YOLOv5 algorithm, using the Adam and 120 optimization algorithm, reaching an accuracy of 58%, validating it then with 30 images, having the averages of 85% accuracy, 91.5 sensitivity and 81.5% F1-Score. Concluding that the algorithm with convolutional neural networks helps to properly identify pavement failures, being important for authorities to do proper maintenance.