Desarrollo de una metodología para estimación de la deforestación mediante el análisis multitemporal de imágenes multiespectrales en un entorno de análisis basado en objetos geográficos (GEOBIA).
Fecha
Autores
Autor corporativo
Título de la revista
ISSN de la revista
Título del volumen
Editor
Compartir
Director
Altmetric
Resumen
The research proposal develops a methodology fully implemented in a geographical object oriented environment (GEOBIA), for the estimation and quantification of the spatiotemporal dynamics of the forest cover (loss or deforestation), by analyzing a time series of medium spatial resolution multispectral images (Landsat). Processes for the proper multispectral and multitemporal data struturation and construction of the time series were formulated, and based on the historical data maps of the Forest and Non-Forest maps (IDEAM) the parameters for the optimization and definition of the feature space, segmentation, training and classification of the time series of objects through four Machine learning algorithms: Vector support machines (SVM), decision trees (DT), Random Forests (RF) and K- Nearest neigbor (KNN) were determined. A methodological framework was formulated, based on the concept of Temporal Link between objects in the time series, for the generation of deforestation objects by means of the temporal analysis of the Forest and Non-Forest objects classified at each moment of the time series; A object oriented accuracy assesment, STEP Matrix, was applied since the geometric and thematic aspects of the generated objects are considered in comparison to the reference map. The results obtained allow us to conclude that a fullyy object - oriented change detection methodology facilitates the modeling of thematic, geometric and temporal characteristics for the analysis of changes in land cover and that the concepts of analysis and spatial modeling of cartographic processes can be implemented in the same time series classification process.