An earthquake damage identification approach from VHR image using mathematical morphology and machine learning

Alatas E. O., Taşkın Kaya G.

NEURAL COMPUTING & APPLICATIONS, cilt.34, sa.21, ss.18757-18771, 2022 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 34 Sayı: 21
  • Basım Tarihi: 2022
  • Doi Numarası: 10.1007/s00521-022-07452-6
  • Dergi Adı: NEURAL COMPUTING & APPLICATIONS
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, PASCAL, Applied Science & Technology Source, Biotechnology Research Abstracts, Compendex, Computer & Applied Sciences, Index Islamicus, INSPEC, zbMATH
  • Sayfa Sayıları: ss.18757-18771
  • Anahtar Kelimeler: Earthquake damage assessment, Classification, Very high resolution image, Mathematical morphology, Attribute filtering, Feature selection, ATTRIBUTE PROFILES, FEATURE-SELECTION, BUILDING DAMAGE, 2003 BAM, RESOLUTION
  • İstanbul Teknik Üniversitesi Adresli: Evet

Özet

An accurate earthquake-induced damage assessment from very high resolution satellite images requires a joint use of spectral and spatial information. The spatial relations between pixels in damaged areas should properly be included in the classification model in order to exploit the spatial information. Morphological Profiles (MPs) and Attribute Profiles (APs) are the state-of-the-art image processing methods that produce a set of filtered images, which are able to highlight specific patterns representing a specific land cover class. In this study, the MPs and the APs are used to create additional spatial features for two different very high resolution post-event satellite images, acquired from the City of Bam in Iran and the City of Port-au-Prince in Haiti. These contextual features are then analyzed by means of a feature selection algorithm, called Minimum Redundancy Maximum Relevance, to find the most relevant features contributing the damage class the most. A final feature subset of selected features is analyzed using two different classifiers, which are k-nearest neighbors and support-vector machines. The results show that the use of a proper configuration of those profiles can significantly improve the classification accuracy and the quality of the thematic map, but generalization of the classification model is limited especially for the larger areas.