Assessing the accuracy of satellite-derived land cover maps classified using a hybrid pixel-based and object-based image analysis technique

The objective of this study is to assess the accuracy of various satellite-derived land cover maps classified using a hybrid pixel-based and object-based image analysis technique. A 2006 QuickBird image covering a part of the eastern side of Tsukuba City, Japan, was used to derive the land cover maps. First, we classified the image using a pixel-based technique. Second, we generated five different sets of object-based segments. And third, we combined the pixel-based classified map and segmentation results. This integration created a hybrid pixel-based and object-based classification procedure. The individual accuracy of the five hybrid classified land cover maps and the pixel-based classified land cover map was assessed using a pixel-based approach. The segments used in the classified land cover map that achieved the highest accuracy based on the pixel-based accuracy assessment approach were further evaluated using a polygon-based accuracy assessment approach.