Title Page
ABSTRACT
초록
Contents
NOMENCLATURE 11
CHAPTER 1. INTRODUCTION 12
CHAPTER 2. PREVIOUS METHODS OF LOCATING CENTER COORDINATES 14
2.1. Circular Hough Transform Analysis 14
2.2. Crystallographic Information Framework Analysis 15
2.3. Rule-based Baseline Model 16
CHAPTER 3. PROPOSED METHOD BASED ON DEEP LEARNING 21
CHAPTER 4. EXPERIMENTAL RESULTS 26
CHAPTER 5. CONCLUSION 33
REFERENCES 34
Table 1. Summarization of each deep learning model tested in this research. 26
Table 2. Numerical value of each ResNet model tested in this research. 28
Table 3. Numbers of layers and parameters for each ResNet model tested in this research. 29
Table 4. Numerical value of each model tested in this research. 30
Figure 1. Circular Hough Transform Analysis: (a) schematic showing accumulator array voting process and (b) schematic showing a method of finding center coordinates... 15
Figure 2. Schematic showing a method of identifying crystal planes by measuring the Euclidean distance between a point on the ring patterns and center coordinates. 16
Figure 3. Flowchart showing algorithmic steps of a rule-based baseline model. 18
Figure 4. Schematics showing RANSAC method used to locate center coordinates. 19
Figure 5. Flowchart showing deep learning object localization technique procedure. 22
Figure 6. Schematics showing steps of convolutional deep learning. 23
Figure 7. Schematics showing process of extracting feature map through convolution. 24
Figure 8. Schematics showing procedure of max pooling which extracts maximum values from each square region of pixel values. 24
Figure 9. Schematics visually and numerically explaining IoU. 27
Figure 10. Schematics visually showing the definition of theta; range covered on diffraction pattern. 27
Figure 11. Graph showing IoU values of each ResNet model tested in this research. 28
Figure 12. Graph showing IoU of each model tested in this research. 29
Figure 13. Distribution plot showing visualizations of each deep learning model with regard to parameter sizes and IoU=0.7 performances. 30
Figure 14. Schematics showing visualizations of the rule-based baseline model and deep learning model IoU results when sufficient data are given. Blue dot is ground truth, red... 31
Figure 15. Schematics showing visualizations of the rule-based baseline model and deep learning model IoU results when insufficient data are given. Blue dot is ground truth,... 31