The road infrastructure conditions are directly related to the safety
and operational cost of transportation. Potholes are defects in the
paving that affect safety on the road. Therefore, identifying potholes
is an important step in defining road maintenance and intervention
strategies. Among the approaches used to detect defects in roads
are vibration techniques, laser scanning and 3D reconstruction, and
finally methods that are vision-based. These vision-based methods
utilize image processing, considered low cost and that can be performed
by common two-dimensional cameras. This research aims
to combine digital image processing and deep learning concepts facilitating
the recognition of pothole-like defects in road images with
asphalt paving. In order to carry out these experiments, different
network architectures were used.