Abstract

Cloud free multispectral scanner (MSS) data of LANDSAT were analysed for studying the effect of the size of training samples on classification accuracy over the following test sites: Sao Jose dos Campos (23° 10' S, 45° 50' W) , Cachoeira Paulista (22° 40' S, 45° W) Jardinopolis (21° S, 47° 50' W).

With the help of ground observations and aerial photography, Sao Jose dos Campos and Cachoeira Paulista were divided into their respective land use classes whereas Jardinopolis was divided into its agricultural classes. LANDSAT MSS data were used to divide each of these classes into two independent data sets -- training fields (about 20 percent of the area) and test fields. Using training fields of each class, test fields were classified using the single-cell option of the Image- 100 as well as a sample classifier based on Bhattacharyya distance. An identical analysis will be done using about ten percent, as well as one percent of the area, for training. This work has many applications since the training phase of the data analysis of earth resources is expensive in the digital as well as the analog mode. If time permits, the effect of the distance of the training samples on the classification accuracy of the test samples will be investigated.

Date of this Version

1977

Share

COinS