Automated classification and thematic mapping of bacterial mats in the North Sea

With the current availability of high quality optical sensors and the advancements of Autonomous Underwater Vehicles (AUVs), it is becoming increasingly accessible to acquire extremely large sets of benthic habitat images. Manual characterization and classification of such large number of images for relevant geological or benthic features can become very difficult and time consuming. This paper presents a novel method for automated segmentation, classification and thematic mapping of bacterial mat from shell chaff and sand, on mosaics created from an image survey on the North Sea. The proposed method uses completed Gabor filter response, grey level co-occurrence matrix (GLCM) and local binary pattern (CLBP) as feature descriptors. After chi-square and Hellinger kernel mapping of feature vector, Probability Density Weighted Mean Distance (PDWMD) is used for classification. Initial segmentation is done using TurboPixels. Our proposed method achieves the highest overall classification accuracy and have moderate execution times compared with the set of methods that are representative of the state-of-the-art in automated classification of seabed images. Our work illustrates that applying automated classification techniques to mosaic composites produces a rapid (in terms of expert annotation time) technique of characterizing benthic areas that can be used to track changes over time.