[Hyperspectral Bambusoideae discrimination based on Mann-Whitney non-parametric test and SVM]

Area of Science:

• Plant Spectroscopy
• Machine Learning in Botany
• Non-parametric Statistical Methods

Background:

• Accurate identification of bamboo species (Bambusoideae) is crucial for ecological and agricultural applications.
• Leaf-level hyperspectral data offers potential for non-invasive species discrimination.
• Previous methods may lack efficiency in identifying optimal spectral features for distinguishing closely related species.

Purpose of the Study:

• To develop and validate methods for discriminating between three bamboo species: MaoZhu, LeiZhu, and XiaoShunZhu.
• To identify optimal spectral bands using non-parametric tests and pattern recognition for accurate species differentiation.
• To evaluate the effectiveness of Support Vector Machine (SVM) algorithms in classifying bamboo species based on selected spectral features.

Main Methods:

• Acquisition of leaf-level hyperspectral data for MaoZhu, LeiZhu, and XiaoShunZhu.
• Application of the Mann-Whitney non-parametric test to extract optimal discriminating spectral bands.
• Classification of bamboo species using the Support Vector Machine (SVM) algorithm within identified optimal bands.

Main Results:

• Optimal discriminating spectral bands were identified for each pairwise comparison of bamboo species, significantly reducing redundant bands.
• The Mann-Whitney test effectively eliminated a substantial percentage of invalid distinguishing bands (30.0% to 57.7%).
• High discrimination accuracies (93.3% to 98.4%) and generalization accuracies (86.7% to 93.3%) were achieved using SVM in the optimal bands.

Conclusions:

• The combination of Mann-Whitney non-parametric tests and SVM provides a valid and accurate approach for bamboo species discrimination using hyperspectral data.
• The identified optimal spectral bands are effective features for distinguishing between MaoZhu, LeiZhu, and XiaoShunZhu.
• This methodology demonstrates strong potential for practical applications in remote sensing and vegetation management for bamboo species identification.