Abstracts Category : Other

Add abstract

Want to add your dissertation abstract to this database? It only takes a minute!

Search abstract

Search for abstracts by subject, author or institution

Share this abstract

Comparison of neural network models applied to size prediction of atmospheric particles based on their two-dimensional light scattering patterns

by Daniel Priori

Institution: Universidade Federal de Santa Catarina
Year: 2017
Posted: 02/01/2018
Record ID: 2166626
Full text PDF: https://repositorio.ufsc.br/xmlui/handle/123456789/181236


Abstract

A obteno do tamanho projetado de partculas atmosfricas prismticas de imensa importncia em diversos aspectos da vida prtica. Partculas expelidas por erupes vulcnicas podem por em risco a aviao civil e militar. Cristais de gelo presentes em nuvens, dependendo de seu tamanho e formato, alteram as propriedades radiantes das nuvens que podem, por sua vez, afetar significativamente os modelos climticos. Uma forma indireta de se obter informaes sobre as partculas prismticas atravs da utilizao de instrumentos que registram padres bidimensionais de disperso de luz. Estas imagens podem ser utilizadas para caracterizar uma partcula cristalina, fornecendo informaes sobre tamanho, razo de proporo, forma, concavidade e rugosidade. Neste trabalho procurou-se aplicar tcnicas de Aprendizado de Mquina, em especial alguns modelos de redes neurais artificiais e tcnicas de anlise de dados, de forma a encontrar um modelo que apresente um desempenho satisfatrio na tarefa de predio do tamanho projetado das partculas cristalinas. Os modelos de redes neurais testados foram do tipo Feed Forward Multi-Layer Perceptron com regularizao Bayesiana, as redes neurais do tipo Funo de Base Radial, e as redes Deep Learning do tipo Autoencoders, a qual tambm foi aplicada com o propsito de reduo dimensional. Tambm foram testadas as tcnicas de anlise de dados de reduo dimensional utilizando Anlise de Componentes Principais e invarincia rotao das imagens atravs da Transformada Rpida de Fourier. Os modelos apresentados foram aplicados a uma srie de imagens e seus resultados comparados e analisados. O modelo desenvolvido que utiliza conceitos de Deep Learning com tcnicas de Autoencoder foi aquele que obteve os melhores resultados (performance de 0.9914), em especial na predio de tamanho projetado para as partculas menores, as quais tiveram maiores dificuldades de predio nos outros modelos propostos nesse trabalho.; Abstract : Obtaining the projected size of atmospheric prismatic particles is of immense importance in many aspects of practical life. Particles expelled by volcanic eruptions may threat to civil and military aviation. Ice crystals present in clouds, depending on their size and shape, can modify the radiant properties of clouds that can significantly affect the climate models. An indirect way of obtaining information on prismatic particles is through the use of instruments that record two-dimensional light scattering patterns. These images can be used to characterize a crystalline particle, providing information on size, aspect ratio, shape, concavity and roughness. In this work we tried to apply Machine Learning techniques, especially some models of artificial neural networks and techniques of data analysis, in order to find a model that presents a satisfactory performance in the task of predicting the projected size of the crystalline particles. The models of neural networks tested were Feed Forward Multi-Layer Perceptron neural network with Bayesian regularization, Radial BasisAdvisors/Committee Members: Roisenberg, Mauro (advisor), Sousa, Giseli de (advisor).

Add abstract

Want to add your dissertation abstract to this database? It only takes a minute!

Search abstract

Search for abstracts by subject, author or institution

Share this abstract

Featured Books

Book cover thumbnail image
Electric Cooperative Managers' Strategies to Enhan...
by White, Michael Edward
   
Book cover thumbnail image
Bullied! Coping with Workplace Bullying
by Gattis, Vanessa M.
   
Book cover thumbnail image
The Filipina-South Floridian International Interne... Agency, Culture, and Paradox
by Haley, Pamela S.
   
Book cover thumbnail image
Solution or Stalemate? Peace Process in Turkey, 2009-2013
by Yurtbay, Baturay
   
Book cover thumbnail image
Performance, Managerial Skill, and Factor Exposure...
by Avci, S. Burcu
   
Book cover thumbnail image
The Deritualization of Death Toward a Practical Theology of Caregiving for the ...
by Gibson, Charles Lynn
   
Book cover thumbnail image
Emotional Intelligence and Leadership Styles Exploring the Relationship between Emotional Intel...
by Olagundoye, Eniola O.
   
Book cover thumbnail image
Commodification of Sexual Labor Contribution of Internet Communities to Prostituti...
by Young, Jeffrey R.
   
Book cover thumbnail image
The Census of Warm Debris Disks in the Solar Neigh...
by Patel, Rahul I.
   
Book cover thumbnail image
Risk Factors and Business Models Understanding the Five Forces of Entrepreneurial R...
by Miles, D. Anthony