Land use studies in drylands: an evaluation of object-oriented classification of very high resolution panchromatic imagery

Object-oriented classification approaches offer an alternative to per-pixel methods for assessment of land use and land cover. Combining object-oriented approaches with very high resolution imagery may provide enhanced possibilities for applications requiring land use and land cover data. The aim of this study is to evaluate the application of object-oriented classification of panchromatic very high resolution data in African drylands, where sizes and shapes of fields are varied, and intercropping practised, which might lead to difficulties in image segmentation. The results show that region-based segmentation is sensitive to the proportion of spectral and shape information and the best results were gained when the segmentation was based on predominately spectral information. The accuracy (Kappa value of 0.6) for the object-oriented classification was significantly higher than that for per-pixel classification. However, both the segmentation and the classification were time-consuming based on a trial and error process.