Separación automática de áreas de bosque nativo por aproximación temática a través de la acumulación multitemporal de clases con Índice SAVI bajo.

FRANCISCO DARIO MALDONADO, WALTER FABIAN SIONE, FERNANDO RAÚL TENTOR, PABLO GILBERTO ACEÑOLAZA

Santos

Simposio; XVII Simposio Brasilero de Sensoramiento Remoto; 2017

Instituto Nacional de Pesquisas Espaciais (INPE)

Abstract. The objective of this work was develop an automatic system for discrimination of areas of native forest, agricultural and cattle use without forest. This objective is aimed at the automatic generation of forest mask for input data to an automatic system for change detection and monitoring of native forest, trough successive masking of annual masks of clear cuts. In this work was used a methodological approach based on the separately process of small spatial subsets (administrative districts), that can be processed by simple techniques increasing de accuracy with each methodological iteraction. The first simple technique was "Kmean" classification, the second is the SAVI vegetation index, and then, the apply of thematic label to the classes based on the average of SAVI values for the pixels in class and accumulation of occurrences of "SAVI low" in a frequency map. As a result, the class "zero frecuency" was labeles as "native forest" for compose the ?Map of native forest? and ?no forest?. The results show two classes with low fragmentation, and the "no forest class" with well defined geometric forms. The "forest class" show little fragmentation, despite the frequency map was the result of one serie of independent thematic transformations.