Research on the Method of Imperfect Wheat Grain Recognition Utilizing Hyperspectral Imaging Technology

Area of Science:

• Agricultural Science
• Image Processing
• Machine Learning

Background:

• Wheat is a primary grain crop in China, vital for agriculture and food security.
• Imperfect wheat grains significantly degrade overall quality and impact food security.
• Current detection methods for imperfect grains are often destructive or inefficient.

Purpose of the Study:

• To develop a fast, non-destructive method for identifying perfect and imperfect wheat grains using hyperspectral imaging.
• To compare the performance of different machine learning models for wheat grain classification.
• To enhance wheat quality assessment and contribute to food security.

Main Methods:

• Collected visible near-infrared hyperspectral data from 2100 wheat grains across seven types.
• Applied Savitzky-Golay and Successive Projections algorithms for data preprocessing and dimensionality reduction, selecting 33 effective spectral datapoints.
• Utilized Principal Component Analysis to identify optimal wavelengths (647.57 nm, 591.78 nm, 568.36 nm) and optimized Support Vector Machines, Convolutional Neural Network, and MobileNet V2 models using Particle Swarm Optimization.

Main Results:

• The optimized Support Vector Machines, Convolutional Neural Network, and MobileNet V2 models achieved comprehensive recognition rates of 93.71%, 95.14%, and 97.71%, respectively.
• The MobileNet V2 model demonstrated superior recognition efficiency compared to other tested models.
• Hyperspectral imaging combined with classification models accurately identified imperfect wheat grains.

Conclusions:

• Hyperspectral imaging offers a promising non-destructive approach for identifying imperfect wheat grains.
• The MobileNet V2 model provides high accuracy and efficiency for this classification task.
• This technology can significantly contribute to improving wheat quality control and ensuring food security.