Image classification and exploratory spatial data analysis techniques

Guidelines are needed to select and parameterize analytic techniques, such as classification procedures, for examining remotely sensed data. The two objectives guiding this paper were to: i) establish the assumptions and information requirements about the spatial structure of remotely sensed images made by image classification routines; and ii) design a framework of spatial statistical techniques capable of providing these data. The classification process examined was stratified into two stages, training data compilation and application of classification decision rules. Relevant literature was reviewed for the various procedures in each stage to establish their: spatial assumptions; spatial parameters; the significance of these data; and how they are justified. Several potential guidelines for implementing classification projects were extracted from the results. These guidelines use spatial analytic techniques to decide on the training scheme and classification decision rule to use, along with their parameters. A preliminary sequence of spatial analysis techniques capable of providing this information was compiled. Establishing a more detailed understanding of an image´s spatial characteristics ensures: i) the most appropriate classification routine and parameters may be selected; and ii) classification results may be interpreted more effectively