LAUSR

ABSTRACT The purpose of this paper is to develop a routine to semi-automatically identify the free span, condition when a pipe segment is not supported by the seabed, from the images of the Synthetic Aperture Sonar and Digital Surface Model (DSM) from an Autonomous Underwater Vehicle (AUV) hydrographic survey. Currently, the free span is performed manually in sonar images, with slow and error-prone process, as it depends on human experience and attention. The methodology developed in this paper consists of extracting profiles of pipeline cross sections from the DSM, and then obtaining geometric information about each profile that will be modeled based on Artificial Neural Network (ANN) and Random Forest (RF) for free span condition classification. The ANN and RF allow evaluating and modeling the relationship between the variables extracted in the profile for the classification of free span condition. The results show that the ANN and RF provide satisfactory results for the free span condition classification with global accuracy of 86.8% and 89.9%, respectively.