방사선 조사 밀 종자의 초분광 이미징 및 머신러닝 기반 특성 분석
Characterization of Radiation-Treated Wheat Seeds Using Hyperspectral Imaging and Machine Learning
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Hyperspectral imaging (HSI) is a non-destructive technique capable of detecting seed coatpigmentation and stress-induced alterations in crop seeds. Here, HSI (400-1000 nm) was applied to wheat seeds ofthree genotypes—Keumkang (white coat), a colored wheat line (CW), and a gamma ray-induced mutant (Crimson)—to characterize spectral differences and evaluate responses to irradiation. Raw reflectance spectra were analyzedwithout preprocessing using principal component analysis (PCA) and three machine learning models: SupportVector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGB). Seed coat analysis showedthat Keumkang displayed the highest reflectance, particularly in the visible range (500-700 nm), whereas CW andCrimson exhibited lower reflectance due to pigment accumulation. PCA confirmed clear separation among thegenotypes, while a common absorption near 900 nm suggested links to seed moisture and compositional traits. Underirradiation (chronic gamma, acute gamma, electron beam; 100-500 Gy), PCA revealed distinct, source-dependentpatterns: chronic irradiation induced separation from controls even at low doses, acute irradiation showed limitedseparation, and electron beam responses were weak except at higher doses. Across conditions, SVM consistentlyachieved the highest accuracy (up to 0.89), and key wavelengths repeatedly clustered within 400-500 and 900-1000nm. These results demonstrate that HSI can reliably detect both pigmentation traits and irradiation-induced changes,even without spectral preprocessing. The approach offers methodological simplicity and practical utility for seedquality evaluation, mutation breeding, and agricultural monitoring.
Abstract
Hyperspectral imaging (HSI) is a non-destructive technique capable of detecting seed coatpigmentation and stress-induced alterations in crop seeds. Here, HSI (400-1000 nm) was applied to wheat seeds ofthree genotypes—Keumkang (white coat), a colored wheat line (CW), and a gamma ray-induced mutant (Crimson)—to characterize spectral differences and evaluate responses to irradiation. Raw reflectance spectra were analyzedwithout preprocessing using principal component analysis (PCA) and three machine learning models: SupportVector Machine (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGB). Seed coat analysis showedthat Keumkang displayed the highest reflectance, particularly in the visible range (500-700 nm), whereas CW andCrimson exhibited lower reflectance due to pigment accumulation. PCA confirmed clear separation among thegenotypes, while a common absorption near 900 nm suggested links to seed moisture and compositional traits. Underirradiation (chronic gamma, acute gamma, electron beam; 100-500 Gy), PCA revealed distinct, source-dependentpatterns: chronic irradiation induced separation from controls even at low doses, acute irradiation showed limitedseparation, and electron beam responses were weak except at higher doses. Across conditions, SVM consistentlyachieved the highest accuracy (up to 0.89), and key wavelengths repeatedly clustered within 400-500 and 900-1000nm. These results demonstrate that HSI can reliably detect both pigmentation traits and irradiation-induced changes,even without spectral preprocessing. The approach offers methodological simplicity and practical utility for seedquality evaluation, mutation breeding, and agricultural monitoring.