Classification of poplar stand areas by high-resolution satellite images

Abstract

This work concerns the classification of different crown cover classes of Poplar stands, using high spatial resolution images (Ikonos and Quickbird satellites), in order to provide poplar monitoring. The test sites are two agricultural areas, located in the alluvial plain of northern Italy, close to Alessandria. In order to enhance spectral differences among classes, textural and high-pass filters were applied and vegetation indices (ratio, difference and normalized difference) were processed. Images were then classified by means of an object-oriented approach which include a segmentation process followed by the application of a Standard Nearest Neighbor classifier on different data sets of spectral images (mean and standard deviation images) and shape indices (shape, compactness). The data sets were defined using the Feature Space Optimization tool available in the Definiens®Developer7 software. From a set of attributes, this tool selects the best combination that produces the largest separability among the classes. The shape of the polygons matched the agricultural plots and the classification results were compared with the reference map defined by means of aerial photo interpretation and ground surveys. New poplar classes were defined in order to improve classification results. The accuracy values obtained were satisfactory (close to 73% for Ikonos and 82% for Quickbird images) and they constitute a basis for automated recognition of poplar plantations and for updating poplar stands assessments.

Keywords

• Satellite images
• Very high spatial resolution
• Object-oriented classification
• Vegetation indices
• Shape indices
• Poplar cultivation