Please use this identifier to cite or link to this item:
Title: Machine learning for benthic sand and maerl classification and coverage estimation in coastal areas around the Maltese Islands
Authors: Gauci, Adam
Deidun, Alan
Abela, John
Zarb Adami, Kristian
Keywords: Machine learning -- Development
Image processing
Ocean bottom ecology -- Malta
Decision trees
Issue Date: 2016
Publisher: Universidad Nacional Autonoma de Mexico. Centro de Ciencias Aplicadas y Desarrollo Tecnologico
Citation: Gauci, A., Deidun, A., Abela, J., & Adami, K. Z. (2016). Machine learning for benthic sand and maerl classification and coverage estimation in coastal areas around the Maltese Islands. Journal of Applied Research and Technology, 14(5), 338-344.
Abstract: Analysis of the seabed composition over a large spatial scale is an interesting yet very challenging task. Apart from the field work involved, hours of video footage captured by cameras mounted on Remote Operated Vehicles (ROVs) have to be reviewed by an expert in order to classify the seabed topology and to identify potential anthropogenic impacts on sensitive benthic assemblages. Apart from being time consuming, such work is highly subjective and through visual inspection alone, a quantitative analysis is highly unlikely to be made. This study investigates the applicability of various Machine Learning techniques for the automatic classification of the seabed into maerl and sand regions from recorded ROV footage. ROV data collected from depths ranging between 50 m and 140 m and at 9.5 km from the northeast coastline of the Maltese Islands, is processed. Through the application of the presented technique, 5.23 GB of data corresponding to 2 h and 24 min of footage which was collected during June 2013, was initially cleaned and classified. An estimate for the percentage cover of the two benthic habitats (sandy seabed and maerl) was also computed by using artifacts encountered during the ROV survey and of known dimensions as a reference. Unlike other automatic seabed mapping techniques, the presented prototype processes video footage captured by a down-facing camera and not through acoustic backscatter. Image data is easier and much cheaper to capture. Promising results that indicate a very good degree of agreement between the true and predicted habitat type distribution values, were obtained.
Appears in Collections:Scholarly Works - FacSciGeo
Scholarly Works - FacSciPhy
Scholarly Works - InsSSA

Files in This Item:
File Description SizeFormat 
Machine_Learning_for_benthic_sand_and_maerl_classi.pdf2.43 MBAdobe PDFView/Open

Items in OAR@UM are protected by copyright, with all rights reserved, unless otherwise indicated.