On the Effective Spatial Information for the Multispectral Classification
スポンサーリンク
概要
- 論文の詳細を見る
This paper describes the effective spatial information for increasing the classification accuracy in case of using SPOT HRV data. In this study, we deal with the spatial information, such as the texture feature, the enhancement image and the restoration image. The enhancement image was made with Laplacian operator, and the restoration image with MTF (Modulation Transfer Function). Nine cases, in combination with the spectral data and the spatial information, were executed.<BR>The results of this study are as follows :<BR>1) It was found that the enhancement image and the restoration image are more useful than the texture feature for increasing the classification accuracy.<BR>2) The enhancement image made with Laplacian operator considering the direction of 45 degrees is the most useful.<BR>3) In this case, the Probability of Correct Classification (PCC) and the Division Accuracy are increased respectively from 88.1% to 92.2% and 91.0% to 96.5%. The Error Ratio considering Omission and Commission errors is decreased from 19.0% to 7.5%.<BR>4) In comparison with the operation time (including CPU time), it takes respectively more than half an hour to make the texture feature (500×500 (pixels)) and the restoration image with MTF, but it takes only about 30 seconds to make the enhancement image with Laplacian operator.<BR>The efficiency of Laplacian operator is higher than the operator of making the texture image and the restoration image with MTF.
- 社団法人 日本リモートセンシング学会の論文