A Comparison between the Isodata and the Ecognition Classification Methods on Basis of Field Data

Image analysis algorithms carry out the precise filtering of the information registered on imageries and the sorting out of the surface objects. They provide the capability to automatically recognize similarities and discriminate among different surface objects. The detail and success of discrimination that can be achieved by the use of these classification algorithms constitutes one of the limiting factors for the effective usage of remote sensing products. In the frame of the presented work two different classification concepts are compared, the pixel based and the object oriented classification method. Both start with a segmentation of the image and are followed by a classification. As representative of the former one was selected the ISODATA clustering method and for the latter one the new innovative image analysis method of DELPHI 2 eCognition. The level of the success of the two methods is evaluated by the use of ground truth data. The better precision and the improvement factor in pattern recognition that can be accomplished with the object based classification method are demonstrated. Ecognition’s capability to finer assess already identified-classified surfaces is presented. Moreover, the advantages of the object oriented classification method are discussed and analyzed. The focus of this work is on agricultural matters and particularly on precision farming applications. A way is presented of how the eCognition method could give thrust to remote sensing application for precision farming purposes.