표제지
초록
목차
제1장 서론 10
제1절 연구의 배경 및 필요성 10
제2절 연구 목적 12
제3절 용어 정의 12
제2장 연구사 16
제1절 근적외선 품질분류 16
제2절 붉가시나무 17
제3장 연구 방법 19
제1절 연구재료 19
제2절 충해종자 분류 21
제3절 종자 전처리 22
제4절 NIR 파장 측정 23
제5절 데이터 분석 26
제1항 데이터 전처리 26
제2항 다변량 분석기법 적용 28
제3항 모델 성능 평가지표 30
제4장 연구 결과 32
제1절 NIR 파장의 형태 분석 32
제2절 충해종자 검정 모델 간 비교 33
제5장 고찰 49
제1절 주성분분석 상에서 이상치 처리 49
제2절 파장변수대에 대한 고려 50
제3절 전처리 및 다변량 분석기법의 처리 효과 52
제4절 연구의 의의 및 한계 54
제5절 후속 연구에 대한 제언 55
제6장 결론 57
Abstract 59
참고문헌 61
Table 1. Number of acorns collected from each family 20
Table 2. Evaluation of accuracy and precision 31
Table 3. Evaluation index for raw spectra 35
Table 4. Evaluation index for standard normal variate processed spectra 38
Table 5. Evaluation index for multiplicative scatter correction processed spectra 40
Table 6. Evaluation index for 1st derivative processed spectra[이미지참조] 42
Table 7. Evaluation index for Savitsky-Golay filter processed spectra 44
Figure 1. Location of the seedling seed orchard of Quercus acuta in Jeju 20
Figure 2. Picture of the caterpillar of Curculio dentipes (left), acorn with a small hole (middle) and acorn with an exit hole (left) 21
Figure 3. Water bath treatment for temperature management 22
Figure 4. Acorn spectroscopy using integrating sphere. ①: Lip, ②: Integrating sphere, ③: Light source, ④: Light from the source, ⑤:... 25
Figure 5. Whole VIS-NIR raw spectra from 400 to 1000nm for acorns of Quercus acuta 32
Figure 6. Average of wavelength between sound and damage acorns from 400 to 1000nm 33
Figure 7. Principal component analysis for sound and damaged acorns in wave range from 780 to 1000nm 34
Figure 8. Spectra processed with standard normal variate 400~1000nm (upper) and 780~1000nm (lower) 36
Figure 9. Spectra processed with multiplicative scatter correction at 780~1000nm 39
Figure 10. Spectra processed with 1st derivative 400~1000nm (upper) and 780~1000nm (lower) 41
Figure 11. Spectra processed with Savitsky-Golay filter, 400~1000nm(upper) and 780~1000nm (lower) 43
Figure 12. Top 5 evaluated combinations of analyzation methods, accuracy (a), precision (b), f1 score (c), G mean (d), SG: Savitsky-Golay,... 45
Figure 13. 400~1000nm Savitsky-Golay filter (SG) processed model's PC score plot, t1: PLS principal component that take the largest part... 46
Figure 14. Y-predicted plot, it shows how y predicted values from PLS regression are set to classes by a certain threshold (Y=0.5). Sample... 47
Figure 15. Variable Important for the Projection: A normalized index that explains which spectra in the model were responsible for Y-pre-... 48
Figure 16. Regression Coefficient: Coefficients of the regression equations formulated with components based on PLS, OPLS-DA. Re-... 48
Figure 17. Schematic of outlier detection, outlier removal in 400~1000nm PCA with Hoteling's T² (confidence interval=95%) (up-... 49
Figure 18. PLS weight of 400~1000nm Savitsky-Golay filtered PLS-DA(SGWP) model's 1st PC[이미지참조] 51
Figure 19. Specificity by preprocessing methods, W: 400~1000nm Inf: 780~1000nm P: PLS-DA O: OPLS-DA 53