Fusion of multi-spectral SPOT-5 images and very high resolution texture information extracted from digital orthophotos for automatic classification of complex Alpine areas

  • Authors:

  • Claudio Mariz;Damiano Gianelle;Lorenzo Bruzzone;Loris Vescovo

  • Affiliations:

  • Centro di Ecologia Alpina, Fondazione E. Mach, Viote del Monte Bondone, Trento, Italy,Department of Information Engineering and Computer Science, University of Trento, Via Sommarive, Trento, Italy;Centro di Ecologia Alpina, Fondazione E. Mach, Viote del Monte Bondone, Trento, Italy;Department of Information Engineering and Computer Science, University of Trento, Via Sommarive, Trento, Italy;Centro di Ecologia Alpina, Fondazione E. Mach, Viote del Monte Bondone, Trento, Italy

  • Venue:
  • International Journal of Remote Sensing
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

In areas with complex three-dimensional features, slope and aspect interact with light conditions and significantly affect the spatial structure of images acquired by remote sensing instruments (for example, by changing the distribution of shadows and affecting the texture of high resolution imagery). In this scenario, this paper analyses the potential and the effectiveness of an automatic classification system to identify three fundamental vegetation classes (forest, grassland and crops) in the complex topography of the Italian Alps (Autonomous Province of Trento, Italy). This classification system is based on the fusion of spectral information provided by the SPOT-5 multi-spectral channels (Ground Instantaneous Field of View, GIFOV, equal to 10 m) and textural information extracted from airborne digital orthophotos (GIFOV equal to 1 m) and is designed to be user-friendly. The texture of the digital orthophotos was modelled using defined bidirectional variograms, thereby extracting additional information unavailable in first-order texture analyses. Using SPOT-5 multi-spectral information alone, the classification accuracy in the investigated alpine area was equal to 87.5%, but increased to 92.1% when texture information was included. In particular, the texture information significantly increased the classification accuracy for crops (from 68.9% to 87.9%), especially orchards that tend to be classified as lowland deciduous forests, and herbaceous crops (such as maize) that are often misclassified as grasslands. A further simple majority analysis increased the ability of detecting grassland, crops and urban zones. The combination of the majority analysis and the proposed automatic classification system seems an effective approach to classifying vegetation types in highly fragmented and complex Alpine landscapes on a regional scale.