Aplicação de diferentes métodos de classificação supervisionada de imagem Landsat-5/TM na identificação de cana-de-açúcar

Abstract

The sugar-cane, from the family species Saccharum officinarum is grown in tropical climates, especially in areas where the seasons are well defined (dry winter and rainy summer). This agriculture is of great importance for the country's economy, Brazil is the world's largest producer of that crop. However, sugar-cane cultivation has favorable characteristics for identification in satellite images because it is a semi-perennial crop, grown in large areas. The objective of this work was to evaluate the performance of supervised classifiers for identifying the culture of sugar-cane using satellite images of Landsat-5 sensor Thematic Mapper (TM). The study area is located northwest from the city of Maracajú-MS, Brazil. We propose a suitable method of classification and image processing to map where there is the cultivation of sugar-cane. Treatments were made to restore the image with spatial resolution of 15 meters and radiometric correction+NDVI. In the rankings, we used the Maxver-ICM algorithm and Bhattacharya. The different pre-processing and classifiers applied were subjected to statistical validation using parameters Kappa and overall accuracy. The results indicated a significant potential for supervised classifiers in the identification of sugar-cane. It was concluded that it is possible to obtain accuracies qualified as very good when used the Maximum Likelihood-ICM classifier in both methods of treatment.