A Multispectral Classification using nPDF(n-Probability Density Function) Features as a Spatial Information.
スポンサーリンク
概要
- 論文の詳細を見る
The nPDF (n-dimensional Probability Density Function) is an algorithm for displaying, analyzing and classifying data (H. Cetin, 1990). The nPDF algorithm is useful for multi-dimensional data transformation and reduction. Furthermore the nPDF plots provide a clear perspective of the data distributions. In this study, we newly define nPDF features as the spatial information. This paper discusses whether the classification accuracy is improved or not with Maximum Likelihood Classification (MLC) in combination with the spectral information and the nPDF features. Three cases, for TM and HRV data respectively, were executed as follows : 1) Only using the spectral information, 2) Using nPDF features in addition to the spectral information, and 3) Using the enhancement image with Laplacian operator considering the direction of 45 degrees which was reported as one of the useful spatial information for MLC (T. Tiyip, 1991)<BR>The results of this study are as follows :<BR>1) In case of using nPDF features, the improvement of the classification accuracy is confirmed for both TM and HRV data.<BR>2) Furthermore, it was found that nPDF features as a spatial information is more useful for improving the classification accuracy than the enhancement image.<BR>3) As a result of measurement of CPU time (computing time by using IBM 9121 : Type 320), it takes only about 36 seconds to make the nPDF feature (500×500 pixels). The transaction of data reduction and transformation using nPDF algorithm is fast and dose not require much memory for calculation ; accordingly the practicality should be assured for multi-spectral classification using nPDF features as a spatial information.
- 社団法人 日本リモートセンシング学会の論文