Analysis of high-resolution remote sensing imagery with textures derived from single pixel objects

摘要

The application of co-occurrence matrices for the calculation of contrast in satellite imagery is a common approach. The textural as well as contextual information from these grey level co-occurrence matrix (GLCM) calculations encounter restrictions due to compromises in their practical implementation. As an alternative, a contrast calculation inside an object-based (OBIA) environment (eCognition) using single-pixel objects is considered. This requires fewer compromises in the implementation, with the flexibility of experimenting on the influence of much larger contextual information for single pixels by expanding the search radius. The contextual information based on contrast can be applied in the classification of the agricultural domain as well as a variety of classes in the 1:25.000 land use/cover classification. The OBIA environment enables a rapid evaluation on various spatial and spectral feature attributes. This allows an evaluation of context on an ever increasing search radius without using larger disk space for synthetic imagery. After the initial evaluation, a small selection of essential contrast maps can be exported as GeoTiff files to allow an input for automated methods. If proven useful, GeoTiff export becomes redundant and the integration of classification methods such as self-organizing maps into the OBIA environment allows effective use of contrast characteristics on small and large neighborhoods.