Ginkgo biloba Sex Identification Methods Using Hyperspectral Imaging and Machine Learning.

Area of Science:

• Plant Science
• Remote Sensing
• Machine Learning

Background:

• Ginkgo biloba L. is a valuable, rare dioecious species cultivated worldwide.
• Accurate sex determination is crucial for Ginkgo cultivation and ecological management.
• Existing methods for sex determination can be time-consuming and labor-intensive.

Purpose of the Study:

• To develop a rapid and effective method for determining the sex of Ginkgo biloba using hyperspectral imaging.
• To establish a standard technique framework for sex classification in dioecious plants.

Main Methods:

• Hyperspectral imaging of green and yellow Ginkgo leaves at different growth stages.
• Development of classification models using RGB images, spectral features, and fused spectral-image features.
• Application of machine learning algorithms including ResNet101, Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), and Subspace Discriminant Analysis (SDA).
• Proposal of a two-stage Period-Predetermined (PP) method for enhanced classification.

Main Results:

• ResNet101 achieved 90.27% accuracy on RGB data.
• Machine learning models showed high prediction accuracies: SVM and SDA (87.35%) for green leaves, LDA (98.85%) for yellow leaves.
• The fused spectral-image dataset with the PP method reached an overall accuracy of 96.30% on the prediction set.

Conclusions:

• Hyperspectral imaging provides an efficient and accurate method for Ginkgo biloba sex classification.
• The developed technique framework offers a standardized approach for industrial and ecological applications.
• This method has potential for classifying the sex of other dioecious plant species.