Feature extraction for NDVI AVHRR/NOAA time series classification

One of the biggest problems of agribusiness in Brazil is related to estimation and forecasting of agricultural crops. In this problem, time series classification enters as a way to help production estimation. In this paper, we are concerned with the development of an automatic classifier that identifies the areas covered with the sugarcane culture by using Normalized Difference Vegetation Index (NDVI) time series, from the AVHRR/NOAA data warehouse of Center of Meteorological and Climatic Research Applied to Agriculture (CEPAGRI). We assumed that a multidimensional space generated by information obtained in the harmonics is a appropriate space to study the similarity between time series. Here we used the word features of a series to refer the coefficients extracted by time series in Fourier decomposition. The proposed methodology has shown to be efficient with a high success rate for the classification of the culture of sugarcane in images from Jaboticabal city, in Brazil, 2004/2005.