Neural-Network Based Algorithm for Algae Detection in Automatic Inspection of Underwater Pipelines

Edgar Medina, Mariane Rembold Petraglia, José Gabriel Rodríguez Carneiro Gomes

Abstract: Automatic inspection of underwater pipelines has been a growing challenge for the detection and classification of events, most often performed by Remotely Operated Vehicles (ROVs) and Autonomous Underwater Vehicles (AUVs). This article describes an algorithm for algae detection in underwater pipelines. The algorithm comprises a neural network and a wavelet-based feature extractor. Statistical parameters of the wavelet coefficients that take into account an appropriate algae texture description, as well as the neural network architecture that results in the optimal classifier performance, are selected. A post-processing algorithm, based on clustering of neighboring detection positions, was implemented to enhance the system response. The success rate of the resulting neural network classifier is 93.60%. When compared to support-vector machines (SVMs), the proposed classifier presents similar performance with the advantage of running significantly faster.