Intertidal beach classification in infrared images

• Published in: Journal of coastal research Vol. 70; no. sp1; pp. 657 - 662
• Main Authors: Hoonhout, Bas, Baart, Fedor, van Thiel de Vries, Jaap
• Format: Journal Article
• Language: English
• Published: Fort Lauderdale Coastal Education and Research Foundation 01-04-2014; Coastal Education and Research Foundation, Inc; Allen Press Inc
• Subjects: Aeolian transport; Algorithms; Automated; Automation; Beaches; Classification; Classification (sedimentation); Coastal; Coastal imagery; Coastal research; Coastal structures; Coasts; Conditional random fields; CRF; Data analysis; Datasets; Digital imaging; Environmental monitoring; Feature extraction; features; Image classification; Image processing; Image resolution; Image segmentation; Imagery; In situ measurement; infrared; Infrared imagery; intertidal beach; Machine learning; Mathematical models; Meteorological parameters; Moisture content; Pixels; Sediments; segmentation; Statistical variance; superpixels; Vegetation
• ISSN: 0749-0208; 1551-5036
• DOI: 10.2112/SI70-111.1

Hoonhout, B.M., Baart, F., Van Thiel de Vries, J.S.M. 2014. Intertidal Beach Classification in Infrared Images. In: Green, A.N. and Cooper, J.A.G. (eds.), Proceedings 13th International Coastal Symposium (Durban, South Africa), Journal of Coastal Research, Special Issue No. 70, pp. 657–662, ISSN 0749-0208. Digital imagery is a powerful data source for coastal monitoring, maintenance and research. It provides high-resolution measurements in both time and space. The size and resolution of long-term imagery datasets provide great opportunities, but also pose problems of tractability in the data analysis. In order to fully use the possibilities of these datasets, reliable and automated classification of images is essential. This paper discusses an automated classification approach based on Conditional Random Fields (CRF). The algorithm is applied in pixel space only. Therefore it does not rely on in-situ measurements, nor is there a need for image rectification. The algorithm consists of three steps: segmentation, feature extraction and model training and prediction. We applied the method to a coastal thermal infrared image stream that monitors the wetting and drying of the upper intertidal beach in relation to tide and meteorological parameters. Classification of the upper intertidal beach provides information on the potential sources of Aeolian sediment. The use of 62 extracted features and structured learning proves to provide significantly better classification results compared to algorithms solely based on intrinsic intensity features.