Investigation of classification accuracy of vegetable field in mountainous area with spectral and textural aspects derived from the Landsat-ETM

This study attempts to discriminate vegetable field in mountainous area with spectral and textural aspects derived from Landsat-ETM, a case study in west Java, Indonesia. Vegetable types over the study area may mainly include temperate climate vegetable such as potatoes, cabbages, carrots and tomatoes. Main data used in the study are Landsat-ETM and geographic data. To minimize the effect of illumination differences on the surface reflectance in Landsat-ETM, spectral bands were normalized by digital elevation model (DEM). To test the dimensionality of the TM image, a principal component analysis (PCA) was used. Textural feature was calculated using a grey level co-occurrence matrix (GLCM). Textural features of contrast, mean, entropy and energy were applied. Finally, principal components and textural features were combined to classify land cover types using maximum likelihood classification (MLC). Generally, combination of spectral and texture aspects enhanced classification accuracy of vegetable field compared to use of spectral information only. While the accuracy was unchanged when using energy texture, except for window size of 3×3. The highest classification accuracy of 85.55% was achieved when using contrast (15×15), mean (15×15), entropy (5×5) and energy (3×3), together.