Spelling suggestions: "subject:"modelos dde filamento dde fused"" "subject:"modelos dde filamento dde fuses""
1 |
Avalia??o do m?todo de similaridade dos perfis e de redes neurais artificiais na estima??o do volume de ?rvores / Evaluation of profiles similarity method and artificial neural networks in the estimation of the volume of treesMurta J?nior, Leonidas Soares 19 July 2013 (has links)
Submitted by Rodrigo Martins Cruz (rodrigo.cruz@ufvjm.edu.br) on 2015-01-07T10:46:05Z
No. of bitstreams: 2
leonidas_soares_murta-junior.pdf: 3383554 bytes, checksum: 710b5a08cef75b6338efb0b57b9f6258 (MD5)
license_rdf: 23898 bytes, checksum: e363e809996cf46ada20da1accfcd9c7 (MD5) / Approved for entry into archive by Rodrigo Martins Cruz (rodrigo.cruz@ufvjm.edu.br) on 2015-01-07T10:46:32Z (GMT) No. of bitstreams: 2
leonidas_soares_murta-junior.pdf: 3383554 bytes, checksum: 710b5a08cef75b6338efb0b57b9f6258 (MD5)
license_rdf: 23898 bytes, checksum: e363e809996cf46ada20da1accfcd9c7 (MD5) / Approved for entry into archive by Rodrigo Martins Cruz (rodrigo.cruz@ufvjm.edu.br) on 2015-01-07T10:46:50Z (GMT) No. of bitstreams: 2
leonidas_soares_murta-junior.pdf: 3383554 bytes, checksum: 710b5a08cef75b6338efb0b57b9f6258 (MD5)
license_rdf: 23898 bytes, checksum: e363e809996cf46ada20da1accfcd9c7 (MD5) / Made available in DSpace on 2015-01-07T10:46:50Z (GMT). No. of bitstreams: 2
leonidas_soares_murta-junior.pdf: 3383554 bytes, checksum: 710b5a08cef75b6338efb0b57b9f6258 (MD5)
license_rdf: 23898 bytes, checksum: e363e809996cf46ada20da1accfcd9c7 (MD5)
Previous issue date: 2013 / Coordena??o de Aperfei?oamento de Pessoal de N?vel Superior (CAPES) / O objetivo deste trabalho foi testar procedimentos para determinar o volume individual de ?rvores, consistindo em m?todos alternativos ao processo de cubagem. No cap?tulo 1, foi testado o m?todo da similaridade de perfis combinado com diferentes medidas de similaridade, modelos de taper, m?todos de estima??o de par?metros representativos de estratos sem equa??o e diferentes m?todos de gera??o de equa??o volum?trica. Os dados utilizados foram de 3.620 ?rvores abatidas e cubadas, divididas 62 estratos. Foram medidas as vari?veis DAP, altura total, e os di?metros ao longo do fuste nas posi??es de 0,00; 0,50; 1,00; 1,50 e 2,00 m, e a partir deste, as se??es foram medidas de 2,0 em 2,0 m, at? a altura total. O modelo volum?trico de Schumacher e Hall foi ajustado por estrato. Os modelos de taper utilizados foram o de Kozak, de Garcia, de Ormerod e o Polin?mio de Quinto Grau. As dist?ncias testadas foram a Dist?ncia Euclidiana, Dist?ncia Euclidiana M?dia, Dist?ncia Euclidiana Ponderada e Dist?ncia Euclidiana Quadr?tica. Foram testados 2 m?todos de estima??o dos par?metros dos modelos de afilamento (3 ?rvores; e os par?metros da ?rvore mais pr?xima ao di?metro m?dio) e 3 m?todos de determina??o da equa??o volum?trica (equa??o do estrato mais similar, equa??o das 30 ?rvores mais similares e equa??o gerada com o n?mero de ?rvores referentes a 5 ?rvores por classe de di?metro do estrato sem equa??o definida), que combinados configuraram os 5 procedimentos testados. As estat?sticas utilizadas para verificar a qualidade das estimativas foram bias, raiz quadrada do erro m?dio, correla??o e an?lise gr?fica de res?duos. A valida??o foi realizada com dados independentes para 4 estratos. Os resultados indicaram que n?o houve diferen?a entre as medidas de similaridade utilizadas. Os melhores modelos foram o Polin?mio de Quinto Grau e o de Kozak, sendo o segundo mais indicado devido a sua facilidade de ajuste. O m?todo de gera??o da equa??o baseada nas 30 ?rvores mais similares proporcionou as melhores estimativas volum?tricas. No cap?tulo 2, foi utilizado redes neurais artificiais (RNA) para estimar o volume individual de ?rvores. Os dados foram os mesmos utilizados no cap?tulo 1. As vari?veis de entrada utilizadas foram: entrada 1 (DAP, altura total, di?metros a 0,00; 0,50; 1,00; 1,50 e 2,00 metros de altura); entrada 2 (adi??o do volume at? 2 metros); entrada 3 (adi??o do grau de esbeltez) e entrada 4 (adi??o da vari?vel categ?rica representando a forma do fuste at? 2 metros de altura). Foram retidas as 5 melhores redes por conjunto de vari?veis de entrada. Os dados foram divididos em 60% para treinamento, 20% para teste e 20% para generaliza??o. Para avaliar as estimativas foram utilizadas as estat?sticas bias, raiz quadrada do erro m?dio, correla??o e an?lise gr?fica dos res?duos. A melhor metodologia para gerar as redes foi aplicada a dados independentes para se avaliar a qualidade das estimativas de volume. As redes que proporcionaram as melhores estimativas foram as geradas pelo conjunto de vari?veis de entrada 2 (DAP, altura total, di?metros a 0,00; 0,50; 1,00; 1,50; 2,00 metros de altura e volume at? 2 metros). / Disserta??o (Mestrado) ? Programa de P?s-Gradua??o em Ci?ncia Florestal, Universidade Federal dos Vales do Jequitinhonha e Mucuri, 2013. / ABSTRACT
The objective of this work was to test procedures to determine the volume of individual trees, consisting of alternative methods to process of scaling. In the Chapter 1, was tested the profiles similarity method combined with different measures of similarity, taper models, parameter estimation methods representative of groups without equation and different generation methods of volumetric equations. The data used were 3.620 felled and cubed trees, divided in 62 groups. The variables measured were DBH, total height, and diameters along the stem in positions 0.00; 0.50; 1.00; 1.50 e 2.00 meters, and to this, the sections were measured from 2.0 to 2.0 meters, up to the total height. The volumetric model of Schumacher and Hall was adjusted by group. Taper models used were Kozak model, Garcia model, Ormerod model and The Fiftieth Degree Polynomial. The distances tested were Euclidean Distance, Mean Euclidean Distance, Weighted Euclidean Distance, and Quadratic Euclidean Distance. Two parameter estimation methods from to taper models were tested (3 trees; and the parameters of the tree nearest to the mean diameter) and three generation methods of volumetric equations (equation similar group, equation of 30 most similar trees and equation generated from the number of trees respect to 5 trees per diameter class for group without defined equation), the combination of them, results in five procedures tested. The statistics used to check the quality of the estimates were bias, root mean square error, correlation coefficient and graphic residual analysis. The validation was performed using independent data for 4 groups. The results indicated that there was no difference between the similarity measures used. The best models were the Fifth Degree Polynomial and Kozak, being the second most indicated due to its ease of adjust. The generation method of the equation based on the 30 most similar trees provided the best volumetric estimates. In the Chapter 2, was used artificial neural networks to estimate the volume of individual trees. The data were the same as used in Chapter 1. The input variables were used: input 1 (DBH, total height, diameters at 0.00; 0.50; 1.00; 1.50 e 2.00 meters); input 2 (addition volume of up to 2 meters); input 3 (addition of slenderness degree) and input 4 (addition of categorical variable representing the form of the stem up to 2 meters). The 5 best networks by set of input variables were retained. The data were divided into 60% for training, 20% for testing and 20% for generalization of the network. To evaluate the estimates were used the statistical bias, root mean square error, correlation coefficient and graphic residual analysis. The best method for generating neural network was applied to independent data to evaluate the quality of the volume estimates. The neural networks provided the best estimates were generated by the set of input variables 2 (DBH, total height, diameter at 0.00, 0.50, 1.00, 1.50, 2.00 meters in height and volume to 2 meters).
|
Page generated in 0.1172 seconds