Image analysis to estimate the fractal dimension of soot aggregates
(2018) PHYM01 20172Combustion Physics
Department of Physics
- Abstract (Swedish)
- Examensarbetet har utvecklat en metod för att estimera den fraktala dimensionen från Transmissions Elektron Mikroskop (TEM) bilder av sot aggregat som skapats med hjälp av en mini-CAST sotgenerator. Det är ett underbestämt problem att estimera den fraktala dimensionen, en 3D egenskap, från 2D bilder. Lösningen är syntetiserade TEM bilder som baserats på projektioner av numeriskt simulerade aggregat med specifika fraktala dimensioner. Bilderna användes för att kalibrera en maskininlärningsmetod med värdet från en fraktal bildanalysmetod, kallad Box Counting dimensionen, som fraktalfeature. De skapade kalibreringarna visar en stark korrelation mellan Box Counting dimensionen och den fraktala dimensionen. Detta verifierar att Box Counting... (More)
- Examensarbetet har utvecklat en metod för att estimera den fraktala dimensionen från Transmissions Elektron Mikroskop (TEM) bilder av sot aggregat som skapats med hjälp av en mini-CAST sotgenerator. Det är ett underbestämt problem att estimera den fraktala dimensionen, en 3D egenskap, från 2D bilder. Lösningen är syntetiserade TEM bilder som baserats på projektioner av numeriskt simulerade aggregat med specifika fraktala dimensioner. Bilderna användes för att kalibrera en maskininlärningsmetod med värdet från en fraktal bildanalysmetod, kallad Box Counting dimensionen, som fraktalfeature. De skapade kalibreringarna visar en stark korrelation mellan Box Counting dimensionen och den fraktala dimensionen. Detta verifierar att Box Counting dimensionen, som feature, extraherar de fraktala egenskaperna från en TEM bild. Två olika versioner av Box Counting, kallade standard-Box Counting och random-Box Counting med skillnad i hur boxarna genereras, har testats och jämförts. Random-Box Counting metoden visade sig ge den mest pålitliga kalibreringen och de mest pålitliga estimeringarna. Denna kalibrering är lik en motsvarande kalibrering som skapats med hjälp av samma metod i ett tidigare arbete. 55 TEM bilder av sot som samplats från den kalla avgasluften av en mini-CAST sotgenerator har analyserats. En förbehandling av varje bild följdes av en estimering av den fraktala dimensionen där kalibreringarna från maskininlärningen användes. Det resulterande medelvärdet av alla estimerade fraktala dimensioner från random-Box Counting metoden var 1.94. De evaluerade fraktala dimensionerna i detta arbete är lite högre jämfört med studier som samplat sot direkt i flammor och möjliga orsaker diskuteras. (Less)
- Abstract
- In this master’s thesis a method for estimating the fractal dimension from Transmission Electron Microscope (TEM) images of soot aggregates has been developed, and also applied to TEM images of soot sampled from a mini-CAST soot generator. The problem to estimate the fractal dimension, a 3D property, from a 2D image is undetermined. To solve this, TEM images were synthesised based on projections of numerically calculated soot aggregates with specified fractal dimensions. The images were used to calibrate a machine learning approach with the output of a fractal image analysis method, called the Box Counting dimension, as fractal feature. The created calibrations show a strong correlation between the Box Counting dimension and the fractal... (More)
- In this master’s thesis a method for estimating the fractal dimension from Transmission Electron Microscope (TEM) images of soot aggregates has been developed, and also applied to TEM images of soot sampled from a mini-CAST soot generator. The problem to estimate the fractal dimension, a 3D property, from a 2D image is undetermined. To solve this, TEM images were synthesised based on projections of numerically calculated soot aggregates with specified fractal dimensions. The images were used to calibrate a machine learning approach with the output of a fractal image analysis method, called the Box Counting dimension, as fractal feature. The created calibrations show a strong correlation between the Box Counting dimension and the fractal dimension. This
verify that the Box Counting dimension as a feature extracts the fractal properties from a TEM image. Two different versions of Box Counting, named standard-Box Counting and random-Box Counting with the difference in how the boxes are generated, have been tested and compared. The random-Box Counting method was found to give the most reliable calibration and estimations. This calibration is similar to a corresponding one produced with the same method in previous work. 55 TEM images of soot sampled from the cold gas exhaust of a mini-CAST soot generator were analysed. A preprocessing algorithm was applied on each image followed by an estimation of the fractal dimension where the calibrations were used. The resulting mean of all fractal dimension estimations from the random-Box Counting method was 1.94. The evaluated fractal dimensions in this work is somewhat higher compared to studies probing flame soot and possible reasons are discussed. (Less) - Popular Abstract
- Soot are extremely small particles with sizes of only a few hundred nano meters. Analysing their shape needs a visualisation of some sort where images from an electron microscope has been used. An electron microscope does not only magnify your interest, but also very small soot particles to objects seen with the naked eye. Together with such images an image analysis method was developed to estimate the shape of soot. The drawbacks of this approach is mainly connected to the use of 2D images to estimate a 3D property, which is an undetermined problem. Though results are still promising since the project confirms the approach of using electron microscope images of soot to estimate its shape.
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/8936472
- author
- Roth, Adrian LU
- supervisor
- organization
- course
- PHYM01 20172
- year
- 2018
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- soot, aggregates, fractal, dimension, image, analysis
- language
- English
- id
- 8936472
- alternative location
- http://adrianroth.se/resume/docs/masters_thesis.se
- date added to LUP
- 2018-02-26 16:33:49
- date last changed
- 2018-02-26 16:33:49
@misc{8936472,
abstract = {{In this master’s thesis a method for estimating the fractal dimension from Transmission Electron Microscope (TEM) images of soot aggregates has been developed, and also applied to TEM images of soot sampled from a mini-CAST soot generator. The problem to estimate the fractal dimension, a 3D property, from a 2D image is undetermined. To solve this, TEM images were synthesised based on projections of numerically calculated soot aggregates with specified fractal dimensions. The images were used to calibrate a machine learning approach with the output of a fractal image analysis method, called the Box Counting dimension, as fractal feature. The created calibrations show a strong correlation between the Box Counting dimension and the fractal dimension. This
verify that the Box Counting dimension as a feature extracts the fractal properties from a TEM image. Two different versions of Box Counting, named standard-Box Counting and random-Box Counting with the difference in how the boxes are generated, have been tested and compared. The random-Box Counting method was found to give the most reliable calibration and estimations. This calibration is similar to a corresponding one produced with the same method in previous work. 55 TEM images of soot sampled from the cold gas exhaust of a mini-CAST soot generator were analysed. A preprocessing algorithm was applied on each image followed by an estimation of the fractal dimension where the calibrations were used. The resulting mean of all fractal dimension estimations from the random-Box Counting method was 1.94. The evaluated fractal dimensions in this work is somewhat higher compared to studies probing flame soot and possible reasons are discussed.}},
author = {{Roth, Adrian}},
language = {{eng}},
note = {{Student Paper}},
title = {{Image analysis to estimate the fractal dimension of soot aggregates}},
year = {{2018}},
}