Along with dissemination of high-resolution satellite imagery, object-based classification providing interpretation-like results has attracted attention over the years. Object-based classification is accomplished by two steps, i.e. segmentation and classification. Segmentation is a process to divide pixels into homogeneous spatial group, and classification is a process to determine class of the group based on their feature quantity. It is major advantage that various feature quantities representing texture and shape are available for classification, however, segmentation prior to classification is also important to produce appropriate land use map. In this study, we proposed the manner to assign parameters for segmentation and discussed on feature quantities in the object-based classification to obtain agricultural land features, i.e. 1) field boundaries, 2) growth uniformity, and 3) sowing/planting way, represented by QuickBird image observed in plantation area in Sumatra, Indonesia.Segmentation was defined by three parameters.
“shape” defines homogeneity of object in term of morphological features and
“compactness” is in irregularity.
“scale parameter” defines the size of object. First of all, 0.1, 0.5, and 0.9 were respectively assigned to
“shape” and
“compactness”. Through the nine pair's trials, the best combination was selected by visual comparison, then
determined the
“scale parameter”.
By the method, several heuristics for segmentation were obtained to classify road features representing field boundary and growing variability patch. Following the segmentation assigned the parameter by the manner, classification with spectral, texture, and morphological features were examined. By the results, classification accuracy was improved than the case of unadjusted parameters that was assigned default values for multi spectral data in software, and 1) field boundaries and 2) growth uniformity were acceptably classified.
View full abstract