Identification of beach hydromorphological patterns/forms through image classification techniques applied to remotely sensed data

Abstract

The evaluation of beach hydromorphological behaviour and its classification is a complex issue. The main objective of this study was to develop new methodologies to identify coastal features/patterns. Pixel-based and object-oriented classification algorithms were used and a new approach was developed based on Principal Components Analysis and Histogram (PCAH) segmentation, to identify and analyse morphological features and hydrodynamic patterns. The PCAH method consists of three stages: preprocessing, PCA and histogram-based segmentation. Both manual and automatic approaches were addressed regarding the identification of the classes obtained from the segmentation stage. The dataset was composed from two aerial photographs and one IKONOS-2 image. The supervised classification algorithms present good results for both aerial photographs and the IKONOS-2 image. For the two aerial photographs the best results were found for the maximum likelihood classifier and for the IKONOS-2 image the best result was achieved with the parallelepiped classifier. The object-oriented classification performance for the aerial photographs and for the IKONOS-2 image also presented good results. The PCAH method led to promising results, with proportions of correctly classified pixels greater than 90% for the classes ‘Sea’, ‘Sediments+breaking zone’ and ‘Beach’.

Acknowledgements

We thank the Laboratório Nacional de Engenharia Civil (LNEC), Dra. A. Fonseca for her assistance in the object-oriented classification stage, and the ESA for providing the IKONOS-2 image. J. Pais-Barbosa thanks Fundação para a Ciência e a Tecnologia, Portugal, for financial support (SFRH/BPD/44929/2008).