Hyperspectral Imaging and Machine Learning as a Nondestructive Method for Proso Millet Seed Detection and...

Area of Science:

• Agricultural Science
• Spectroscopy
• Machine Learning

Background:

• Millet (Panicum miliaceum L.) is a valuable cereal crop with diverse cultivars requiring efficient sorting methods.
• Current cultivar classification techniques are often subjective, destructive, and time-consuming.
• Developing nondestructive methods for proso millet cultivar discrimination is crucial for industry stakeholders.

Purpose of the Study:

• To evaluate the feasibility of near-infrared (NIR) hyperspectral imaging for classifying proso millet cultivars.
• To develop and validate a rapid, nondestructive method for distinguishing between ten popular US proso millet cultivars.

Main Methods:

• Investigated 5000 proso millet seeds from ten cultivars using NIR hyperspectral imaging (900-1700 nm).
• Applied principal component analysis (PCA) to reduce data dimensionality, utilizing the first two principal components as spectral features.
• Employed a Gradient Tree Boosting ensemble machine learning algorithm for classification.

Main Results:

• NIR hyperspectral imaging achieved up to 99% accuracy in classifying proso millet cultivars.
• Classification accuracy remained high (98.14% and 97.6%) even with reduced spectral features (15 and 5 wavelengths).
• Principal component analysis effectively reduced data dimensionality while retaining critical spectral information.

Conclusions:

• NIR hyperspectral imaging is a viable, rapid, and nondestructive technique for proso millet seed classification.
• This method offers significant advantages over traditional subjective and destructive classification approaches.
• The findings support the practical application of NIR hyperspectral imaging in the agricultural industry for quality control and seed sorting.