都市域の土地被覆分類におけるピクセルベース手法とオブジェクトベース手法の比較

—高解像度デジタル航空写真を用いて—

抄録

In this study, we compared pixel-based image analysis and object-based image analysis (OBIA) as methods of land cover classification of urban areas, using high resolution digital aerial photography. The study area was Setagaya Ward, Tokyo, Japan, and we carried out supervised classification using aerial photographs with 25-cm spatial resolution, and with both visible bands and a near infrared band. The overall accuracy of the object-based classification was approximately 6 to 20 percentage points higher than that of the pixel-based classification. Both methods tended to over-classify water areas and bare land, specifically, with shadows of buildings and roads in impervious areas tending to be misclassified as water areas and as bare land, respectively. The tendency of over-classification was remarkable in the pixel-based classification, and most of the classified area was a minute area of 1 pixel to several pixels. To evaluate such minute areas (salt-and-pepper effect) in the classification, we calculated the join-count statistics, a kind of spatial autocorrelation index. The tendency of grouping of the object-based classification was stronger than of the pixel-based classification, indicating that the minute areas in the object-based classification were fewer than those in the pixel-based classification. We also calculated the green coverage ratio based on the present classification results, and compared it to that obtained by the municipality of Setagaya Ward. The green coverage ratio from the object-based classification was between the two values by the municipality, whereas the value from the pixel-based classification was smaller than both values. In conclusion, the object-based method is applicable to the land cover classification and extraction of vegetation using high-resolution digital aerial photography in urban areas.