Qualidade física de um solo degradado em recuperação via redes neurais artificiais
Carregando...
Data
2020-03-23
Autores
Chitero, José Guilherme Marques
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
Os solos sob um manejo inadequado, têm suas qualidades físicas, químicas e biológicas afetadas negativamente, dando origem a sua degradação. No estado de São Paulo, grande parte das pastagens estão degradadas e/ou em degradação. Entender como um funciona um solo degradado e sua resiliência e iniciar seu processo de restauração, são fundamentais para desenvolver técnicas de manejo adequado do solo. Inúmeras técnicas estão sendo utilizadas para recuperação de solo degradado, descobrir e detalhar os índices físicos do solo ajuda em como proceder, e com qual técnica de recuperação utilizar; visando isso, este projeto teve por objetivo desenvolver um programa interativo (analisar e classificar) com a utilização das redes neurais artificiais (RNA) para estimar os níveis de recuperação do solo (recuperado (R), parcialmente recuperado (PR) e não recuperado (NR) em função dos atributos físicos e comparar com os dados obtidos via estatística convencional. O experimento foi realizado na Agência Paulista de Tecnologias dos Agronegócios – APTA do Extremo Oeste, no município de Andradina/SP no período de 2015 a 2017, em solo classificado como Argissolo Vermelho Amarelo cultivado com pastagem de Urochloa, com diferentes formas de introdução de Estilosantes cv. Campo Grande (Stylosanthes capitata e S. macrocephala). Os atributos do solo estudados foram: densidade do solo, porosidade do solo (macroporosidade e microporosidade), resistência mecânica a penetração, infiltração de água no solo e diâmetro médio ponderado, nas camadas do solo: 0-10; 0,10-0,20 e 0,20-0,40 m. Os dados foram avaliados pela estatística convencional (teste de Tukey a 5% de probabilidade) e via redes neurais artificiais. O programa foi desenvolvido no ambiente MATLAB e a simulação foirealizada por meio de uma interface gráfica. A RNA que foi utilizada neste trabalho é a Perceptron multicamadas (MLP), composta por três camadas, de entrada, intermediária e a de saída, com algoritmo de treinamento retropropagação (treinamento supervisionado). A camada de entrada é composta por 6 neurônios (atributos físicos) e a camada de saída por 1 neurônio, o qual tem a função de informar a classificação quanto a recuperação física do solo (R, PR e NR). Dos resultados obtidos pela RNA, verificou-se que a rede alcançou um treinamento adequado, com erro quadrado médio (mean square error) baixo, podendo gerar uma alternativa interessante e automática para a classificação e análise de solos em recuperação. Os resultados foram imprimidos em uma interface gráfica autoexplicativa, com gráficos e metadados dos índices físicos e suas classificações quanto à RNA.
Soils under inadequate management have their physical, chemical and biological qualities negatively affected, giving rise to their degradation. In the state of São Paulo, most of the pastures are degraded and / or in degradation. Understanding how a degraded soil and its resilience works and starting its restoration process are essential to develop proper soil management techniques. Countless techniques are being used to recover degraded soil, discovering and detailing the physical indexes of the soil helps in how to proceed, and with which recovery technique to use; with this in mind, this project has developed an interactive program (analyze and classify) using artificial neural networks (ANN) to estimate soil recovery levels (recovered (R), partially recovered (PR) and not recovered ( NR) as a function of physical attributes and compare with data obtained using conventional statistics.The experiment was carried out at the São Paulo Agribusiness Technologies Agency - APTA in the Far West, in the city of Andradina / SP from 2015 to 2017, on classified soil as Red Yellow Argisol cultivated on Urochloa decumbens pasture in recovery since 2012, with different ways of introducing Estiloantes cv. Campo Grande (Stylosanthes capitata and S. macrocephala). The studied soil attributes are: soil density, soil porosity (macroporosity and microporosity), mechanical resistance to penetration, water infiltration in the soil and weighted average diameter, in soil layers: 0-10; 0.10-0.20 and 0.20-0.40 m. The was evaluated by conventional statistics (Tukey's test at 5% probability) and via artificial neural networks. The program was developed in the MATLAB environment and the simulation was performed through a graphical interface. The ANN that was used in this work is the multilayer Perceptron (MLP), composed of three layers, of input, intermediate and output, with backpropagation training algorithm (supervised training). The input layer is composed by 6 neurons (physical attributes) and the output layer by 1 neuron, which has the function of informing the classification regarding the physical recovery of the soil (R, PR and NR). From the results obtained by RNA, it was found that the network achieved adequate training, with a low mean square error, which could generate an interesting and automatic alternative for the classification and analysis of recovering soils. The results were printed on a selfexplanatory graphical interface, with graphs and metadata of the physical indexes and their classifications regarding ANN.
Soils under inadequate management have their physical, chemical and biological qualities negatively affected, giving rise to their degradation. In the state of São Paulo, most of the pastures are degraded and / or in degradation. Understanding how a degraded soil and its resilience works and starting its restoration process are essential to develop proper soil management techniques. Countless techniques are being used to recover degraded soil, discovering and detailing the physical indexes of the soil helps in how to proceed, and with which recovery technique to use; with this in mind, this project has developed an interactive program (analyze and classify) using artificial neural networks (ANN) to estimate soil recovery levels (recovered (R), partially recovered (PR) and not recovered ( NR) as a function of physical attributes and compare with data obtained using conventional statistics.The experiment was carried out at the São Paulo Agribusiness Technologies Agency - APTA in the Far West, in the city of Andradina / SP from 2015 to 2017, on classified soil as Red Yellow Argisol cultivated on Urochloa decumbens pasture in recovery since 2012, with different ways of introducing Estiloantes cv. Campo Grande (Stylosanthes capitata and S. macrocephala). The studied soil attributes are: soil density, soil porosity (macroporosity and microporosity), mechanical resistance to penetration, water infiltration in the soil and weighted average diameter, in soil layers: 0-10; 0.10-0.20 and 0.20-0.40 m. The was evaluated by conventional statistics (Tukey's test at 5% probability) and via artificial neural networks. The program was developed in the MATLAB environment and the simulation was performed through a graphical interface. The ANN that was used in this work is the multilayer Perceptron (MLP), composed of three layers, of input, intermediate and output, with backpropagation training algorithm (supervised training). The input layer is composed by 6 neurons (physical attributes) and the output layer by 1 neuron, which has the function of informing the classification regarding the physical recovery of the soil (R, PR and NR). From the results obtained by RNA, it was found that the network achieved adequate training, with a low mean square error, which could generate an interesting and automatic alternative for the classification and analysis of recovering soils. The results were printed on a selfexplanatory graphical interface, with graphs and metadata of the physical indexes and their classifications regarding ANN.
Descrição
Palavras-chave
Qualidade física do solo, Inteligência artificial, Atributos físicos, RNA, Soil quality, Computational technology, Physical indexes, ANN