RESUMO O estudo das propriedades físicas e químicas do solo é um procedimento de custo e tempo relativamente elevado. Na busca de alternativas para predizer esses atributos a partir de um número menor de amostras do solo, o uso de Redes Neurais Artificiais (RNA) tem sido apontado como uma técnica computacional com grande capacidade de resolver problemas por meio da experiência, e possuem a capacidade de aquisição e posterior aplicação deste conhecimento. Esse trabalho teve por objetivo utilizar a RNA para estimar os atributos físicos e químicos de solo. Os dados utilizados foram provenientes da análise física e química de solo, coletados em 120 pontos amostrais, os quais foram submetidos à análise descritiva, análise geoestatística, treinamento e análise das RNAs. Na análise geoestatística, para cada atributo do solo, foi verificado o modelo de semivariograma que apresentou melhor ajuste ao modelo experimental, e como método de interpolação foi usada técnica da krigagem ordinária. As RNAs foram treinadas, selecionadas considerando a assertividade no mapeamento dos padrões considerados e utilizadas na estimativa de todos dos atributos de solo. O erro médio de cada estimativa obtida pela técnica da krigagem ordinária foi comparado com o erro médio da estimativa obtida pela RNA e, posteriormente foram comparadas com os valores originais por meio do teste-t de Student. Os resultados mostram que a técnica de RNAs apresenta assertividade compatível à krigagem ordinária. O uso da técnica de RNA apresentou-se promissora para obter estimativas de atributos de solo empregando um número menor de amostras de solo. | ABSTRACT Soil physical and chemical analyses are relatively high-cost and time-consuming procedures. In the search for alternatives to predict these properties from a reduced number of soil samples, the use of Artificial Neural Networks (ANN) has been pointed out as a great computational technique to solve this problem by means of experience. This tool also has the ability to acquire knowledge and then apply it. This study aimed at using ANNs to estimate the physical and chemical properties of soil. The data came from the physical and chemical analysis of 120 sampling points, which were submitted to descriptive analysis, geostatistical analysis, and ANNs training and analysis. In the geostatistical analysis, the semivariogram model that best fitted the experimental variogram was verified for each soil property, and the ordinary kriging was used as an interpolation method. The ANNs were trained and selected based on their assertiveness in the mapping of considered standards, and then used to estimate all soil properties. The mean errors of ordinary kriging estimates were compared to those of ANNs and then compared to the original values using Student's t-Test. The results showed that the ANN had an assertiveness compatible with ordinary kriging. Therefore, such technique is a promising tool to estimate soil properties using a reduced number of soil samples.

ABSTRACT Image analysis combined with machine learning models can be an excellent tool for classification of fava (Phaseolus lunatus L.) genotypes and is a low-cost system. Fava is grown by family farmers, mainly, in the Northeast and South regions of Brazil, presenting economic and social importance. Evaluations to gather information on qualitative and quantitative characters of seeds enable the description and distinction of genotypes, allowing the evaluation of variability of plant species, which is essential in breeding programs. The use of image analysis is a fast and economic tool for obtaining large quantity of information. Machine learning techniques have been developed and implemented in the agricultural sector due to technological advances and increasing use of artificial intelligence, which enables the automatization of several processes. In this context, the objective of this work was to evaluate different machine learning models to classify fava genotypes, using data obtained through image analysis. Images of fava seeds were captured using a table scanner (HP Scanjet 2004), set to true color mode, arranged upside down inside of an aluminum box fully closed during the capture of the images for an adequate illumination and prevention of environmental noises. The K-Nearest Neighbor, Naive Bayes, Linear Discriminant Analysis, Support Vector Machine, Gradient Boosting, Bootstrap Aggregating, Classification and Regression Trees, Random Forest, and C50 models were used for the study. Linear Discriminant Analysis was the model that presented the highest efficiency for classifying the genotypes, with an accuracy of 90%.

ABSTRACT Quantifying soil gas emissions is costly, since it requires specific methodologies and equipment. The objective of this study was to evaluate modeling by nonlinear regression and artificial neural networks (ANN) to estimate CO2 emissions caused by soil managements. CO2 emissions were evaluated in two different soil management systems: no-tillage and minimum tillage. Readings of CO2 flow were carried out by an automated closed system chamber; soil temperature, water content, density, and total organic carbon were also determined. The regression model and the ANN models were adjusted based on the correlation of the variables measured in the areas where the soil was managed with no-tillage and minimum tillage with data of CO2 emission. Artificial neural networks are more accurate to determine correlations between CO2 emissions and soil temperature, water content, density, and organic carbon content than linear regression.