Identificación Inteligente de Cultivos Estivales mediante sensores remotos

Abstract

The agricultural sector has been experiencing a profound change towards digital transformation, seeking to anticipate soil-climatic events that can influence productive results and facilitate remote decision-making. Crop monitoring with remote sensors and the application of machine learning techniques are tools that allow identifying crops, changes in phenology, anomalies and at lower or almost zero cost than traditional methods. Identifying crops, obtaining estimates and yield predictions are essential to improve the country's productive prospects. Currently, in Argentina very few developments in different contexts (John Deere, Monsanto, http://www.sepa.inta.gob.ar, http://www.agropuma.com) have the scientific and technological means to carry out this study using satellite technology and Machine learning techniques and that is within the reach of producers. In this first stage of research, corn was classified vs. soybean in Sentinel-2 multispectral images with data recorded from the 2019/2020 campaign of the General López department (Santa Fe). The data set for classification consisted of the selection of representative bands, the calculation of statistical measurements (average, maximum, minimum, deviation, etc.) and the calculation of vegetative indices (NDVI, SAVI). Two machine learning algorithms Random forest (RF) and Support vector machine (SVM) were implemented with an overall accuracy of 79% and 73%. The general objective is to extend these results to a larger agro-productive agricultural region and to estimate yields and production.