Inteligencia artificial, coberturas del suelo y cultivos de coca en Norte de Santander: Estudio de caso en los Municipios de Tibú, El Tarra y Sardinata

Abstract

This document presents the development of a methodology to determine the possible rural areas with coca crops in the municipalities Tibú, Sardinata and El Tarra in the Norte de Santander department; implementing knowledge of satellite image analysis and Artificial Intelligence, achieving reliable results. To carry out the project, SENTINEL-2 satellite images were used between 2018 and 2020. The images went through pre-processing, and then visually analyzed with the help of the polygon layer of the SIMCI project, which frame the areas with coca crops in the country. Once the areas with coca crops were determined, the samples were extracted in true color composition to consolidate the data set of the "COCA_SI" label. The "COCA_NO" label was consolidated with samples taken from a free data repository, plus samples taken from the municipalities without coca crops in the Boyacá department. In the design stage of the final structure of the Convolutional Network and in the definition of the hyperparameters, a constant bibliographic revision, multiple tests of the code, adjustments, comparison and analysis of the results obtained after each validation and the analysis derived from the confusion matrix, which finally led to choose the model with the best performance, obtaining an accuracy of 98%, showing positive results, so the objective of the project was achieved.