Análisis estructural de baches viales mediante redes neuronales profundas

The rapid growth of the global population has intensified vehicular traffic, posing a significant challenge to its management. In developing countries such as Colombia, traffic accidents exhibit high mortality rates, largely attributed to road defects, such as cracks and potholes. Given this problem, the objective of this study was to develop an automated system for detecting defects in urban pavements using Convolutional Neural Networks (CNNs) to classify 11 types of road surface failure. The methodology involved creating a dataset from images of defective roads, which was used to train deep neural network models. Two optimizers, SGDM and ADAM, were evaluated using color and grayscale pictures, processed in MATLAB® and validated by civil engineering experts. The results showed that the SGDM optimizer achieved an accuracy of 74.67 % with color images, while ADAM achieved a performance of 52.51 % with grayscale images. These findings demonstrated the potential of CNNs and digital image processing techniques to automate pavement inspection, increasing both efficiency and accuracy in evaluating road infrastructure. Finally, it is concluded that the use of deep neural networks represents a viable alternative for developing intelligent pavement management systems and supports the implementation of data-driven solutions to optimize urban road maintenance.