Multiple-point geostatistical simulation for post-processing a remotely sensed land cover classification

摘要: A post-processing method for increasing the accuracy of a remote sensing classification was developed and tested based on theory multiple-point geostatistics. Training images are used to characterise joint variability continuity target spatial pattern, overcoming limitations two-point statistical models. Conditional simulation (MPS) applied land cover derived from remotely sensed image. data were provided in form “hard” (land labels), “soft” constraints (class probability surfaces estimated using soft classification). The MPS compared two alternatives: traditional filtering (also method) contextual Markov random field (MRF) classifier. approach increased relative these alternatives, primarily as result curvilinear classes. Key advantages that, unlike MRF classifier, (i) it incorporates rich model correlation process smoothing spectral (ii) has advantage capturing utilising class-specific training patterns, example, classes with distributions.