Uso de dados geográficos e previsão automatizada com redes neurais na agricultura

Abstract

The world population demands an increased production of food for its subsistence. The search for processes and models that collaborate to increase productivity is a key factor to achieve results that minimize costs to the producer and contribute to decision making about the crop's needs. Geotechnologies, in conjunction with data analysis methods, allow monitoring of the productive area aiming to achieve better results. Thus, this work aims to present functionalities and methodologies that allow obtaining, processing, visualization, and prediction of indexes in agriculture in an automated way, not burdening the final user. In Article 01, it presents the application of tools in the generation of a database to be used by agriculture through an automation process for the implementation of a geographic database structure, using free software PostgreSQL+Postgis, which is fed by relevant data sources for agriculture (limits of the analyzed area and obtained by sensors). Thus, it results in a temporal data visualization tool that includes 16 indexes for analyzing the behavior of the productive area. In Article 02, from the consolidated base, Python language and convolutional neural networks were used to establish predictive models of NDVI indices, reaching mean absolute error (MAE) in the predictions from 0.16 to 0.17, with the presentation of better results for the use of networks that used windows of 5 images prior to the subsequent prediction. Both works made it possible to structure procedures for implementing, structuring, feeding, and analyzing agricultural databases in an automated way, minimizing costs and increasing the agility to generate relevant information for crop development. With this, the methodologies and tools used can be considered by producers or analysts to monitor, perform preventive actions and consequently achieve better results in agricultural production