Preliminary evaluation of eCognition object-based software for cut block delineation and feature extraction

Pixel-based classifications have difficulty adequately or conveniently exploiting expert knowledge or contextual information. Object-based image-processing techniques overcome these difficulties by first segmenting the image into meaningful multipixel objects of various sizes, based on both spectral and spatial characteristics of groups of pixels. The segments (objects) are assigned classes using fuzzy logic and a hierarchical decision key. To date, the main drawback has been the lack of effective software. This paper evaluates a simple object-based approach to classifying forest cut blocks near Revelstoke, British Columbia, in a Landsat enhanced thematic mapper plus (ETM+) image using recently released eCognition software. Three subsets of the image test the software. Two subsets are chosen to evaluate it in different land-cover situations, and the third subset is used to test the effectiveness of replicating a rule system developed elsewhere in the image. Rules for class assignment include traditional spectral signatures, polygon shape parameters, and context with other classes. Use of multiple image object levels (sizes) greatly assists in image classification. Cut block scars, areas of mature, young, and sparse forest, water, and urban features are classified with significantly higher accuracy than using a traditional pixel-based method. The software allows expert knowledge to be documented and retained for use elsewhere or for later modification.