Underwater Image Classification by Neural Networks

In this paper, a vision-based system which help an autonomous underwater vehicle (AUV) to detect and track a given stationary target placed on the sea bottom, for e.g. a sealine, is presented. A Back Propagation neural networks is applied to perform a real-time classification of the input image pixels into two different classes corresponding to sealine edge or other regions. A post-processing method allows to evaluate sealine edges positions also in noisy images. Results on real underwater images representing a sealine in many different situations, e.g., presence of seaweed and sand, different illumination conditions, water depth, etc., are shown.