Clasificación del estado superficial de pavimentos rígidos con imágenes digitales y técnicas de inteligencia artificial

Abstract

Although there are different artificial intelligence methods for the detection and classification of damages to road infrastructure, the absence of research in the management and maintenance of rigid pavements is evident. This research focuses on establishing an artificial intelligence model for the evaluation of deterioration in rigid pavements by estimating the present serviceability index (PSI). An artificial intelligence algorithm was developed that processes image data to obtain a PSI value. This research was approached through 3 axes: 69 videos were captured from different sectors of Bogotá D.C., from which a library of 5046 images of different states of rigid pavement was obtained. These were classified into groups of 200 images by a group of engineers experienced in pavement management and allowed for the training of convolutional neural networks (VGG16, Resnet 50, OWN, EfficientNetB0). The best performance metrics, precision of 75% and loss of 0.68, were obtained from the implementation of the pre-trained EfficientNetB0 network. This work is a starting point for the management and maintenance of rigid pavements with novel technological tools, improving the amount of time and human effort employed daily.