Using patch metrics as validation for contextual classification of heathland vegetation

This article presents a method to assess the accuracy of vegetation maps for which contextual information has been included in the classification process. It is well known that land use classification may benefit from combining spatial and spectral information. Consequently, many classification techniques incorporating spatial information have been implemented. To compare various contextual classification techniques, the shape of vegetation patches, in the spatially enhanced maps, are statistically linked to their counterparts in the spectral classification result, on which these spatial enhancements are applied. To this end, measures for the change in shape of patches are introduced. The shape of any patch is characterized by the edges between the patch and its neighbors. Therefore, patch shape can be represented by an edge map in which each pixel gets the value of the number of classes that are different from the class label of the central pixel in a four-adjacency neighborhood. Rather than defining a single metric for the edge map difference, an error matrix is used to depict not only how many edges have changed with respect to the reference, but also by how much they have changed. The method is tested on contextual classification results of heathland vegetation.

[1]  Josiane Zerubia,et al.  Supervised segmentation of remote sensing images based on a tree-structured MRF model , 2005, IEEE Transactions on Geoscience and Remote Sensing.

[2]  Stephen E. Fienberg,et al.  Discrete Multivariate Analysis: Theory and Practice , 1976 .

[3]  Iphigenia Keramitsoglou,et al.  Habitat mapping using machine learning-extended kernel-based reclassification of an Ikonos satellite image , 2006 .

[4]  Johannes R. Sveinsson,et al.  Spectral and spatial classification of hyperspectral data using SVMs and morphological profiles , 2008, 2007 IEEE International Geoscience and Remote Sensing Symposium.

[5]  Josiane Zerubia,et al.  Bayesian image classification using Markov random fields , 1996, Image Vis. Comput..

[6]  Lorenzo Bruzzone,et al.  A Novel Protocol for Accuracy Assessment in Classification of Very High Resolution Images , 2010, IEEE Transactions on Geoscience and Remote Sensing.

[7]  Charalambos Kontoes,et al.  Kernel based re-classification of Earth observation data for fine scale habitat mapping , 2005 .

[8]  Paul Scheunders,et al.  Spatial hyperspectral image classification by prior segmentation , 2009, 2009 IEEE International Geoscience and Remote Sensing Symposium.

[9]  Donald Geman,et al.  Stochastic Relaxation, Gibbs Distributions, and the Bayesian Restoration of Images , 1984, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[10]  Lorenzo Bruzzone,et al.  Quality assessment of classification and cluster maps without ground truth knowledge , 2005, IEEE Transactions on Geoscience and Remote Sensing.

[11]  Huseyin Gokhan Akcay,et al.  Automatic Detection of Geospatial Objects Using Multiple Hierarchical Segmentations , 2008, IEEE Transactions on Geoscience and Remote Sensing.

[12]  Stan Z. Li,et al.  Markov Random Field Modeling in Image Analysis , 2001, Computer Science Workbench.

[13]  Russell G. Congalton,et al.  Assessing the accuracy of remotely sensed data : principles and practices , 1998 .