Characterizing Chinese saffron Origin, Age and grade using VNlR hyperspectral imaging and Machine learning

Area of Science:

• Agricultural Science
• Analytical Chemistry
• Spectroscopy

Background:

• Saffron (Crocus sativus L.) is a valuable spice and medicinal herb.
• Its economic value is significantly influenced by geographical origin, age, and grade.
• Objective quality assessment is crucial for the saffron industry and consumers.

Purpose of the Study:

• To develop and validate a method for identifying saffron origin, age, and grade.
• To leverage visible-near infrared hyperspectral imaging (VNIR-HSI) combined with machine learning.
• To explore data fusion strategies for enhanced classification accuracy.

Main Methods:

• Saffron samples were graded according to ISO 2011/2010 standards.
• Visible-near infrared hyperspectral imaging (VNIR-HSI) was employed for data acquisition.
• Machine learning algorithms (SVM) coupled with preprocessing methods (MSC, CARS) were utilized.
• Mid-level data fusion of image and spectral features was performed.

Main Results:

• Age was found to have a greater influence on saffron grade than geographical origin.
• MSC-CARS-SVM effectively classified saffron origin.
• FD-CARS-SVM effectively classified saffron age and grade.
• Data fusion improved origin and age prediction accuracies to 98.3% and 97.9%, respectively.
• The spectral FD-CARS-SVM model achieved 89.6% accuracy for grade identification.

Conclusions:

• The proposed VNIR-HSI, machine learning, and data fusion method provides a robust approach for saffron quality characterization.
• This technique offers theoretical basis and technical support for the saffron industry.
• Accurate identification of origin, age, and grade benefits both producers and consumers.