Entropy based noise clustering soft classification method for identification of wheat crop using time series MODIS data

In this study, time series MODIS-Terra MOD13Q1 data have been used for the identification of wheat crop in a test site in the state of Haryana in India. Wheat is the dominating crop in this region having large and homogeneous fields. A total of seven date data have been taken between November 2011 to April 2012, corresponding to the different phenological stages of wheat crop of this study area. The Normalized Difference Vegetation Index (NDVI) and Soil Adjusted Vegetation Index (SAVI) have been generated for each date MODIS data. Further, these indices were then used to make the temporal indices sets. The Transformed Divergence (TD) separability based optimized sets of temporal spectral indices were then used for classification of wheat crop using the supervised entropy based noise clustering soft classification algorithm. For the assessment of accuracy the Receiver Operating Characteristic (ROC) analysis has been used. It is observed that the highest area under ROC curve is found for NDVI `Three´ date temporal dataset combination, which is a combination of sowing, flowering and maturity stages of wheat phenology. Further, the inclusions of milking stage and maturity stage data has decreased the accuracy. Thus, it may be concluded that `Three´ date combination of NDVI generated from MODIS yields best result for identification of wheat crop using entropy based noise clustering classifier.