Simplificando a Interpretação de Laudos de Análise de Solo com Deep Learning em Nuvem

Abstract

Since the soil impacts directly on agricultural productivity, its con-servation through the correct application of nutrients and fertilization is of pa-ramount importance. In this work, we propose a software architecture and amobile application capable of assisting farmers and agronomists in interpretingsoil analyses generated from laboratories. The software architecture was desig-ned for cloud environments and the mobile application is the interface for cap-turing and presenting data. Initially, it was necessary to create a database withdifferent image types and configurations. All images from the dataset were tre-ated to eliminate noise (such as brightness, shadows and distortions) and usedto evaluate two Deep Learning solutions (Google Vision and Tesseract OCR),where Tesseract OCR proved to be more accurate using the same images. Inaddition to offering the mobile application, which is the first step, the researchcarried out reveals several technological deficiencies and opportunities for in-novations in the field of soil science.