Classification Performance Improvement of Agricultural Crops in Multi-temporal Images Using Textural Information in Ghorveh County

Features extracted from single-date satellite images can not lead to high accuracy in crop classification. As a result, in this article use of multi-temporal images and textural information is assessed. This research evaluates discrimination of four crops i.e. alfalfa, wheat, potato and cucumber in Ghorveh county (Kordestan province) with single-date and multi-temporal images (SPOT5 & ASTER). Seven images of four different months were obtained and separability analyses determined optimal scene combination (i.e. July 2nd image for single-date image and July 2nd – October 21st for two-date image) for classification. GLCM method has been used for extracting textural information. Approximate window size was determined with variogram, then extracted features were stacked with single-date and two-date spectral bands. The best single-date image overall accuracy (classification without texture) was 24% higher than the worst single-date image accuracy (76.02% versus 52.28%). Accuracy in two-date image reached to the highest levels (89.61%) and only five-date image has higher accuracy than two-date image, but with adding texture features to five-date image, accuracy decreased 5% lower than two-date image with texture features. This can indicate that texture has more importance than multi-date images. The highest classification accuracy was reached with the best two-date image (97.48%). In single-date images, July 2nd image reached to highest classification accuracy (95.2%). Results indicated using multi-temporal images along with texture features has very higher accuracy in comparison to conventional methods (without texture features or multi-temporal images).