Title Page
ABSTRACT (EN)
ABSTRACT (KR)
Contents
1. INTRODUCTION 20
1.1. Research Background 20
1.1.1. Remote Sensing 22
1.1.2. Spectroscopy 27
1.1.3. Polymer Spectrum 34
1.1.4. Green House Gases Spectrum 37
1.1.5. HITRAN 42
1.1.6. Deep Learning Model 43
1.1.7. Remote Sensing with Deep Learning 47
1.2. Purpose of Research 52
2. MICROPLASTICS DETECTION 53
2.1. EXPERIMENTS 53
2.1.1. Optical Characteristics of Polymer 53
2.1.2. Environmental Effects on Spectrum 55
2.1.3. Polymer Detection Experimental Setup 59
2.1.4. Polymer Spectrum Analysis UI 62
2.2. DATA PREPROCESSING 64
2.2.1. Peak Extraction 64
2.2.2. Convex Hull 64
2.2.3. Gaussian Deconvolution with Levenberg Marquardt 66
2.2.4. Environment Compensation 70
2.3. RESULTS 72
2.3.1. Microplastics Detection 72
2.3.2. Microplastics Detection with Deep Learning 80
3. LOW CONCENTRATED GHGs DETECTION 84
3.1. EXPERIMENTS 84
3.1.1. Green House Gases Optical Characteristics 84
3.1.2. Concentration of Mixed Gases 87
3.1.3. Gas Detection Experimental Setup 89
3.1.4. Gas Data Acquisition UI 94
3.2. DATA PREPROCESSING 95
3.2.1. Convert to Transmission 95
3.2.2. 2D Imageization 101
3.2.3. Sequential Noise Elimination 104
3.3. RESULTS 109
3.3.1. Gas Detection with CNN 110
3.3.2. Gas Detection with LSTM 112
4. CONCLUSIONS AND FUTURE WORK 117
REFERENCES 120
OUTCOMES OF RESEARCH 128
Table 1-1. Molecular interactions associated with the energy frequencies 34
Table 1-2. Vibrational modes for the CH₄ molecules 41
Table 2-1. Representative peak wavelengths for different plastic polymers studied by Garaba's group 55
Table 2-2. NRmax for Representative Peak Wavelength of Plastic Polymer in Various Environment 76
Table 3-1. MFC RS-485 Protocol Commands 92
Table 3-2. Detectable ppm Concentration per each GHGs 116
Figure 1-1. Representative fields of deep learning applied on remote sensing 21
Figure 1-2. Various platforms and sensors used for remote sensing 22
Figure 1-3. Passive sensors (multispectral and hyperspectral) 23
Figure 1-4. Examples of Landsat and Sentinel imagery over North Florida 24
Figure 1-5. Active sensors (LiDAR and Radar) 25
Figure 1-6. Classification of a Quad-Pole Radarsat-2 scene of Flevoland, Holland 26
Figure 1-7. Electromagnetic wave consists of two components: an electric field and a magnetic field 28
Figure 1-8. Electromagnetic spectrum 29
Figure 1-9. Absorption spectroscopy 30
Figure 1-10. Schematic diagram showing (a) The attenuation of radiation passing through a sample and (b) Redefining P0 as the radiant power transmitted by the black[이미지참조] 31
Figure 1-11. Transmittance spectroscopy 32
Figure 1-12. Reflectance spectroscopy 33
Figure 1-13. Illustration of Alphatic C-H from 2.2.4-Dimethyl Pentane (a) The 4th overtone (5v) and (b) The 3rd overtone (4v) with combination bands[이미지참조] 35
Figure 1-14. Illustration of (a) Polyethylene, (b) Polyethylene Oxide, (c) Polypropylene, (d) Polystyrene monomer and (e) Polyarcrylic acid 36
Figure 1-15. The amount of radiation forced by GHGs based on changes in the concentrations in the Earth's atmosphere since 1750 38
Figure 1-16. The average temperature of the Earth and the concentration level of CO₂ in the Earth's atmosphere during the last 1000 years 38
Figure 1-17. Overview of CO₂ absorption spectra at standard pressure and temperature 39
Figure 1-18. HITRAN simulation of absorption spectra for major atmospheric species 42
Figure 1-19. Deep learning model 43
Figure 1-20. Neural network with convolutional layers 44
Figure 1-21. Basic LSTM structure 45
Figure 1-22. Basic GRU structure and gate calculations 46
Figure 1-23. Number of conference papers and articles in the Scopus database for a general search on ['Deep Learning' AND 'Remote Sensing'] 47
Figure 1-24. CNN architecture for the classification of hyperspectral images 48
Figure 1-25. Faster R-CNN adopted for improving performance for SAR ATR 49
Figure 1-26. Analysis for finding land usage by hyperspectral images 50
Figure 2-1. Measured data of (a) Reflection and transmittance of plastics, and (b) Reflection, transmittance and characteristic absorption peaks of plastics measured by... 54
Figure 2-2. Liquid water absorption spectral over a wide wavelength band 56
Figure 2-3. 3-fundamental modes of vibration for water 57
Figure 2-4. Data of (a) Transmission spectra through ice for VIS-NNIR region 58
Figure 2-5. Schematic and pictures of sample materials. (a) Plastic polymers, (b) Plastic polymers exposed of ice, (c) Plastic polymers covered with ice, and (d) Floating plastic... 59
Figure 2-6. Schematic illustration of the technical arrangement of a hyperspectral field or laboratory measurement 60
Figure 2-7. Schematics of experimental diagram (SCC: Signal Conditioning Components, DAQ: Data Acquisition) 62
Figure 2-8. Execution windows of microplastic classification (e.g. the ice is selected in the 'environment') 63
Figure 2-9. Measured data of (a) Raw data with convex line, (b) Subtracted data for baseline rearrangement 65
Figure 2-10. Convex hull procedures (a) Before Pi (b) Processing Pi, and (c) After adding Pi[이미지참조] 66
Figure 2-11. Spectral analysis of PE (a) Divided into multiple gaussian functions, and (b) LM algorithm residual. 69
Figure 2-12. Spectrum analysis of PET in water (a) LM method curve fitting, and (b) water compensated area, and re-identified 71
Figure 2-13. Polymer detection results 72
Figure 2-14. Pure polymer optical characteristics in NIR range 74
Figure 2-15. The spectral of the characteristic peak wavelength of plastic polymers (a) Exposed to ice, (b) Covered with ice, and (c) Floating on water 77
Figure 2-16. Degree of identification and classification of plastics under various conditions (a) Single species and (b) Polymer mixture 79
Figure 2-17. Database for microplastics detection 82
Figure 2-18. CNN model for microplastics detection 83
Figure 2-19. (a) Loss in each epochs (b, c) Test results 83
Figure 3-1. HITRAN transmission database for (a) CH₄ and (b) CO₂ 85
Figure 3-2. Experimental data (a) CH₄ and (b) CO₂ 86
Figure 3-3. Automatic convertion from ppm concentration to SCCM 88
Figure 3-4. SpectraWiz to compare (a) noise and (b) main peaks in multiple signals 89
Figure 3-5. Inner structure of MFC 90
Figure 3-6. Picture of (a) MFC connection and (b) Conventional software 'WIZ-701' 91
Figure 3-7. Experimental setting for low concentration gas detection 93
Figure 3-8. Parallel programming to get multiple sensors data 94
Figure 3-9. Measurement of (a) Detected halogen spectrum in different intensity (b) Derivative of each spectrum 96
Figure 3-10. Spectrum of (a) Linear curve fitted for 1000ppm of CO₂, and (b) Linear curve fitted for 7000ppm of CH₄ 97
Figure 3-11. Rearranged spectrum of (a) 1000ppm of CO₂, and (b) 7000ppm of CH₄ using method 2 98
Figure 3-12. Spectrum of (a) Original signal and filtered using (b) Moving average(window=9), (c) Median filter (window=9), (d) Savitcky-Golay (window=9), (e) Minmax... 99
Figure 3-13. Baseline fitted spectrum of (a) 1000ppm of CO₂, and (b) 7000 ppm of CH₄ 100
Figure 3-14. Organized dataset as 2D image (a) 3000ppm, (b) 5000ppm, and (c) 7000ppm of CH₄ 102
Figure 3-15. Organized dataset as 2D image (a) 10000ppm, (b) 30000ppm, and (c) 50000ppm of CO₂ 103
Figure 3-16. Filtering applied 3000ppm of CH₄ (a) Original and (b) Filtering in wavelength, and (c) Filtering in time 106
Figure 3-17. Randomly extracted image from single original data 108
Figure 3-18. Models attempted to accurately detect GHGs 110
Figure 3-19. CNN model for gas detection 111
Figure 3-20. RNN many to one stacking 112
Figure 3-21. LSTM model for gas detection 113
Figure 3-22. CH₄ 500ppm data processing per range and similarity filtering results with image 114
Figure 3-23. CO₂ 1000ppm data processing per range and similarity filtering results with image 115