A Review of the Effectiveness of Spatial Information used in Urban Land Cover Classification of VHR Imagery

  • Salehi, B.,
  • Zhang, Y.,
  • Zhong, M.,
  • Dey, V.,


Land cover classification of very high resolution (VHR) imagery in urban areas is an extremely challenging task, because of the low intra-class (within-class) and high inter-class (between-classes) spectral similarities of impervious land cover types (such as buildings and traffic areas). Over the past decade, a significant amount of research has been conducted on the incorporation of spatial information along with spectral information of VHR imagery into urban land cover classification. The spatial information includes textural, morphological and contextual measures extracted from VHR imagery, as well as LiDAR- and photogrammetrically-derived DSM and existing GIS data layers. In this paper, a comprehensive review of recent literature was conducted to evaluate the effectiveness of such measures in land cover classification of urban areas using VHR imagery. For each measure, a comprehensive list of papers for both pixel-based and object-based classification is provided. In addition, the classification results of representative publications are reported and its advantages and limitations in both pixel-based and object-based approaches are discussed. It has been found that, in general, object-based classification performs better than pixel-based approaches, since it facilitates the use of spatial measures by segmenting the image. Moreover, utilizing spatial measures significantly improves the classification performance for impervious land cover types, while may have no effect or even lower the classification accuracy for classes of vegetation and water surfaces. Textural measures are more commonly utilized in pixel-based approaches, while morphological measures have better performance in object-based classification. The effect of contextual measures on classification is enhanced when these measures are used in conjunction with two other measures, particularly in object-based approaches. Although ancillary data shows a very high potential to address the problem of spectral-based classifiers in separating spectrally similar impervious land cover types, incorporating such data, particularly photogrammetrically-derived DSM, in classification is still in a very early stage and requires significant exploration and development.