Prediction and soil texture classification by multivariate image analysis
| Autor: | Morais, Pedro Augusto de Oliveira |
|---|---|
| Přispěvatelé: | Oliveira, Anselmo Elcana de, Coelho, Clarimar José, Antoniosi Filho, Nelson Roberto |
| Jazyk: | portugalština |
| Rok vydání: | 2016 |
| Předmět: | |
| Zdroj: | Biblioteca Digital de Teses e Dissertações da UFG Universidade Federal de Goiás (UFG) instacron:UFG |
| Popis: | A textura, ou granulometria, do solo e de nida pela distribui c~ao quantitativa das part culas minerais menores que 2 mm: areia, argila e silte. Esses indicadores f sicos possibilitam a classi ca c~ao do solo e orientam o manejo, a irriga c~ao e a adi c~ao de insumos agr colas. As metodologias usuais para an alise textural s~ao laboriosas, destrutivas, utilizam agentes qu micos oxidantes e essa an alise e bastante requisitada nos laborat orios de an alise da fertilidade do solo. Logo, s~ao imprescind veis investiga c~oes e o desenvolvimento de metodologias alternativas que sejam operacionais e limpas. Nessa dire c~ao, esse estudo prop~oe a utiliza c~ao de an alise multivariada de imagens digitais para previs~ao da textura do solo e classi ca c~ao. Para tanto, 60 amostras de solo diversi cadas foram consideradas para an alise textural pelo m etodo da pipeta e para obten c~ao de imagens digitais no sistema de cor RGB (Red, Green, Blue) em formato Ti . A correla c~ao entre as imagens digitais e os teores de areia, argila e silte for feita por Regress~ao por Quadrados M nimos Parciais (PLS) e por Regress~ao Linear M ultipla associada com o Algoritmo das Proje c~oes Sucessivas (SPA-RLM). Os melhores modelos apresentaram um ndice de acerto de 100%. Portanto, a predi c~ao textural do solo atrav es de imagens e uma t ecnica promissora por ser limpa, barata e operacional. The texture or grain size of the surface are de ned by the quantitative distribution of the mineral particles smaller than 2 mm: sand, clay and silt. These physical indicators enable soil classi cation and guide the management, irrigation and the addition of agricultural inputs. Although the usual methods for textural analysis are laborious and destructive, using chemical oxidizing agents, this kind of analysis is quite required in soil fertility laboratories. Therefore, it is essential to research and develop alternative methodologies that are operational and clean. In this way, this study proposes the use of multivariate analysis of digital images to predict and classify soil texture. For this purpose, 60 samples of diverse soil were considered to textural analysis by the pipette method and for obtaining digital images in color system RGB (Red, Green, Blue) in Ti format. The correlation between digital images and the percentage of sand, clay and silt is made by Partial Least Squares Regression (PLS) and Multiple Linear Regression algorithm associated with the Successive Projections (SPA-MLR). The best models had a 100 % success rate. Therefore, the prediction texture soil through images is a promising technique to be clean, inexpensive and operational. Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|