표제지
초록
Abstract
목차
I. 연구배경 및 목적 19
1. 연구배경 19
가. 갯벌 생태계의 중요성 19
나. 갯벌 생태계 관리, 조사 현황 및 한계점 21
다. 갯벌 생태계 조사 기술 발전의 필요성 23
2. 연구목표 25
3. 연구개요 26
II. 연구방법 29
1. 연구지역 선정 29
가. 태안 신두리 해안 갯벌 29
나. 태안 바람아래 갯벌 32
2. 연구대상 선정 34
가. 태안 신두리 갯벌 대형저서동물 34
나. 태안 바람아래 갯벌 대형저서동물 37
3. 무인항공기 기반 갯벌 대형저서동물 탐지 방법론 개발 39
가. 무인항공기 기반 갯벌 대형저서동물 촬영 및 탐지 방법론 개발 39
나. 무인항공기 기반 갯벌 대형저서동물 탐지 방법론 활용성 평가 53
4. 딥러닝 모델 기반 갯벌 대형저서동물 자동탐지 및 분류 54
가. 무인항공기 기반 딥러닝 모델 데이터 구축 54
나. 딥러닝 모델 정확도 및 활용성 평가 67
III. 연구결과 및 논의 70
1. 무인항공기 기반 갯벌 대형저서동물 탐지 방법론 개발 70
가. 무인항공기 기반 갯벌 대형저서동물 촬영 및 탐지 방법론 개발 70
나. 무인항공기 기반 갯벌 대형저서동물 탐지 방법론 활용성 평가 73
2. 딥러닝 모델 기반 갯벌 대형저서동물 탐지 및 분류 방법론 개발 90
가. 무인항공기 기반 딥러닝 모델 데이터 구축 90
나. 딥러닝 모델 정확도 및 활용성 평가 94
IV. 결론 107
1. 주요 연구 결론 107
가. 무인항공기 기반 갯벌 대형저서동물 탐지 방법론 개발 107
나. 딥러닝 기반 대형저서동물 자동 탐지 및 분류 방법론 개발 108
2. 시사점 110
가. 무인항공기 기반 갯벌 대형저서동물 탐지 방법론 개발 110
나. 딥러닝 기반 대형저서동물 자동 탐지 및 분류 방법론 개발 114
3. 한계점 116
참고 문헌 118
Table 2.1. Research target species in Baramarae tidal flat 38
Table 2.2. DJI Phantom4 and gimbal optical camera specifications 41
Table 2.3. DJI Inspire 2 and Zenmuse X5S specifications 41
Table 2.4. Trimble 8s VRS/RTK-GPS specifications 41
Table 2.5. Hyperparameter for the optimization of the deep learning model (Sindu-ri tidal flat macrobenthos) 62
Table 2.6. Crab types and populations by image set 65
Table 2.7. Hyperparameter for the optimization of the deep learning model (Baramarae tidal flat macrobenthos) 66
Table 2.8. Summary of accuracy metrics to evaluate deep learning 68
Table 3.1. Population of macrobenthos in Baramarae tidal flats based on UAV images 74
Table 3.2. Population of macrobenthos in Baramarae tidal flats based on DSM altitude 76
Table 3.3. Macrobenthos list of 2019 National investigation of marine ecosystem in Baramarae tidal flat 88
Table 3.4. Results of deep learning model accuracy of macrobenthos in Sindu-ri tidal flat(U-net) 95
Table 3.5. Results of deep learning model accuracy of macrobenthos in Baramarae tidal flat(U-net) 97
Table 3.6. Results of deep learning model accuracy of macrobenthos in Baramarae tidal flat(HRnet) 97
Table 3.7. Results of deep learning model accuracy of macrobenthos in Baramarae tidal flat(HRnet-OCR) 98
Table 3.8. Results of deep learning model population and appearance rate of macrobenthos in Baramarae tidal flat 101
Table 3.9. Comparison of the five traditional methods and the UAV, deep learning method to sample intertidal crabs 106
Figure 1.1. Defining coastal ecosystems 20
Figure 1.2. Overall framework of the dissertation contents 28
Figure 2.1. Study area(Sindu-ri tidal flat) 31
Figure 2.2. Study area(Baramarae tidal flat) 33
Figure 2.3. In-situ monitoring using UAV 34
Figure 2.4. Ocypode stimpsoni montoring area(Sindu-ri tidal flat) 35
Figure 2.5. Ocypode stimpsoni and burrows in Sindu-ri tidal flat 36
Figure 2.6. Research target species image in Baramrae tidal flat 38
Figure 2.7. UAV images by flight altitude 44
Figure 2.8. Example of redundancy of UAV photography 46
Figure 2.9. Surveying GCPs locations in Sindu-ri tidal flat and dune 48
Figure 2.10. Example of pixel intensity change as window moves 50
Figure 2.11. Process of SfM 51
Figure 2.12. Sindu-ri coastal image registration using Pix4DMapper 52
Figure 2.13. Process of building deep learning image sets through Ocypode stimpsoni and burrow classification 55
Figure 2.14. Labeling sample for classification 56
Figure 2.15. Classification of artificial neural network-based deep learning image processing 58
Figure 2.16. ReLU function 60
Figure 2.17. U-net architecture 60
Figure 2.18. HR-net architecture 64
Figure 3.1. UAV Images of Ocypode stimpsoni(red) and burrow(yelloew) 71
Figure 3.2. UAV Images of Uca lactea(yellow) and Macrophthalmus japonicus(blue) 72
Figure 3.3. Distribution of macrobenthos and DSM in Baramarae tidal flats 77
Figure 3.4. Distribution of Uca lactea and DSM in Baramarae tidal flats 78
Figure 3.5. Distribution of Uca arcuata and DSM in Baramarae tidal flats 79
Figure 3.6. Distribution of Scopimera globosa and DSM in Baramarae tidal flat 80
Figure 3.7. Distribution of Macrophthalmus japonicus and DSM in Baramarae tidal flats 81
Figure 3.8. Profile analysis results for line of DSM in Baramarae tidal flats and distribution of macrobenthos 83
Figure 3.9. Profile analysis results for line A-B of DSM in Baramarae tidal flats and distribution of macrobenthos 84
Figure 3.10. Profile analysis results for line C-D of DSM in Baramarae tidal flats and distribution of macrobenthos 85
Figure 3.11. Profile analysis results for line E-F of DSM in Baramarae tidal flats and distribution of macrobenthos 86
Figure 3.12. UAV based survey point and national investigation of marine ecosystem survey point in Baramarae tidal flat 89
Figure 3.13. Labeling results of Ocypode stimpsoni(yellow) and burrow(red) 91
Figure 3.14. Labeling results of Uca lactea male(green), Uca lactea female(yellow), and Scopimera globosa(purple) 93
Figure 3.15. Ocypode stimpsoni(optical image, labeled image, and deep learning prediction result) 95
Figure 3.16. Burrow(optical image, labeled image, and deep learning prediction result) 95
Figure 3.17. Orthoimage-based Ocypode stimpsoni and burrow automatic detection and classification results 103
Figure 4.1. Scopimera globosa feeding pellet 111
Figure 4.2. Territorial battles between Uca lactea 111
Figure 4.3. Gobiidae and Mollusca in Baramarae tidal flat 113
Figure 4.4. Result sample of HRnet deep learning 115