A CLASSIFICATION METHOD WITH A SPATIAL-SPECTRAL VARIABILITY

A classification method which takes into account not only spectral inf ormation but also spatial information is proposed for high-spatial-res olution multi-spectral scanner data such as Landsat TM and SPOT HRV da ta. Such a spatial feature can be used with spectral features in a uni fied way, in a pixel-wise Gaussian-based Maximum Likelihood classifica tion (MLC) because the probability density function of a spatial featu re is similar to the normal distribution under some assumptions. From experiments, there was found to be a substantial improvement in the ov erall classification accuracy for TM forestry data. The probability of correct classification (PCC) for the new clearcut and the alpine mead ow classes increased by 7 to 97 per cent correct by adding the spatial feature.