Examination of the optimum segmentation in the object-based image analysis for forest stand type classification

For object-based forest stand type classification using high resolution satellite data, the relation of the parameter of segmentation and classification accuracy was investigated. A target area is forested landscape in the Kirishima area over both Ebino city of Miyazaki and Kirishima city of Kagoshima, the southern part of Kyushu Island. A pan-sharpen IKONOS data (1-m of spatial resolution) was employed in this analysis. In addition to varying the scale parameter from 100 to 1000, some combinations of color and shape criterion were examined. As a result of measuring the accuracy of forest stand type classification, when scale parameter was 300, the highest classification accuracy was achieved. It was indicated that the color criterion also affects classification accuracy in this study. Consequently, it would be greatly concerned with classification accuracy whether target patches are delineated adequately by segmentation, and this paper indicated that quantitatively.