Comparison of Multi-scale Images of an Agricultural Land Using Polygon-based Classification Techniques

Polygon-based classification was performed on multi-scale images of SPOT4 XS, SPOT5 XS, IKONOS XS, QuickBird XS and QuickBird Pansharpaned (PS) covering an agricultural area located in Karacabey, Turkey. The objective of the study was to assess the classification accuracies of different spatial resolution images on an agricultural land using the polygon-based classification techniques. The existing boundaries of the agricultural fields were updated through on screen digitizing within-field boundaries. Polygon-based classification of the images was then carried out using the common bands. The polygon-based classification techniques used include (i) pre-polygon classification, and (ii) post-polygon classification. To perform the pre-polygon classification, for each field, the mean values were calculated. Then, a Maximum Likelihood Classification (MLC) was performed using the mean bands. For the post-polygon classification, first, the images were classified on per-pixel basis using the MLC technique. Then, for each field, the frequencies of the classified pixels were computed and the field was assigned the label of the model class. The assessments of the classification results showed that, for all images used in this study, the post-polygon classification approach provided better results than the pre-polygon classification approach. Of the images used, the 4-m resolution IKONOS XS image provided the highest overall accuracy of 88.6%. On the other hand, the lowest overall classification accuracy was provided by the 20-m resolution SPOT4 XS image. The overall classification accuracies were computed for the QuickBird XS (2.44-m) and QuickBird PS (0.61-m) images as 83.7% and 85.8%, respectively.