An area of 95200 km(2) in central Spain was mapped using IRS-WiFS data to test the potential use of this imagery for regional cover type mapping. In addition to the original WiFS red and NIR spectral bands, textural-contextual images were computed as means and variances of the NDVI and the NIR and red bands over windows of different sizes (3x3 to 50x50 pixels) and included in the analysis. An iterative classification of the imagery was performed using a maximum likelihood classifier with feature selection by spectral separability indices. Results obtained in this study show that IRS-WiFS is a valuable source of information for forest cover mapping at regional scales. Semi-natural areas (comprising forests, shrubs and grasslands) and forests were classified with 93 and 83 mean accuracy, respectively. The classification accuracy of most cover types increased when textural-contextual information computed over large windows (larger than those reported in the literature) were used for classification in combination with spectral data. References: 11
展开▼
