Dimension reduction methods applied to coastline extraction on hyperspectral imagery

ARSLAN O., AKYÜREK Ö., Kaya Ş., Şeker D. Z.

GEOCARTO INTERNATIONAL, vol.35, no.4, pp.376-390, 2020 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 35 Issue: 4
  • Publication Date: 2020
  • Doi Number: 10.1080/10106049.2018.1520920
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Aqualine, Aquatic Science & Fisheries Abstracts (ASFA), Environment Index, Geobase, INSPEC
  • Page Numbers: pp.376-390
  • Keywords: Hyperspectral imagery, coastline, dimension reduction, remote sensing, image analysis, INDEPENDENT COMPONENT ANALYSIS, CLASSIFICATION METHODS
  • Kocaeli University Affiliated: Yes


In this study, dimensionality reduction (DR) methods on a hyperspectral dataset to explore the influence on the process of extraction of coastlines were examined and performance of different DR algorithms on the detection of coastline in Bosphorus, Istanbul was investigated. Among these methods, principal component (PC) analysis, maximum noise fraction and independent component (IC) analysis were used in the experiments with the aim of comparing. The study was carried out using these well-known DR techniques on a real hyperspectral image, an Hyperion data set with 161 bands, in the course of the experiments. Three different classifiers (i.e. ML, SVM and neural network) were used for the classification of dimensionally reduced and original images to detect coastline in the region. The DR results were evaluated quantitatively and visually in order to determine the reduced dimensions of the image subsets. Findings show that there is no significant influence of using DR methods on the dataset on the detection of coastline.