Application of combined pixel-based and spatial and spatial-based approaches for improved mixed vegetation classification using ikonos

Classifying a mosaic of coffee systems, each in a different stage of structural complexity is not obvious when that ranges from monoculture to a complex agro-ecosystem, with various shade and fruit trees mixed in different degrees of density. Distinction into different sub-classes incorporating tree complexity and tree cover, is important as tree density and the generally related amount of litter are important from a soil erosion perspective. In this study, the objective was to classify different coffee garden systems plus several other minor vegetation classes existing in the area using IKONOS in Sumberjaya district, Lampung Province, Indonesia. Pixel-based classification approach was integrated with spatial-based approach to reach an improved classification result. In the supervised pixel-based approach training samples are collected to generate statistical parameters for the classifier to classify the whole image. The spatial-based approach refers to segmentation procedure, known also as object-based classification. Two methods of integration were explored and pure pixel-based-approach was as well conducted for comparison purpose. Results were then tested using ground check data. The methods tested are: pure spectral approach of (a) supervised classification using maximum likelihood classifier, integration with segmentation which was done in two ways, by (b) classifying the segments and by (c) combining the pixel-based classified image with segment image using majority rule. Of all the three methods the combination using majority rule showed the highest overall accuracy. Several points were discussed as feedback to the methods tried as well as to improve the classification result

---

• View source