Machine learning approach for wheat variety identification using single-seed imaging

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate wheat varietal identification is crucial for agriculture but challenging due to cultivar similarity.
• Traditional methods are subjective and labor-intensive, hindering efficiency and reliability.

Purpose of the Study:

• To develop and compare automated wheat varietal classification frameworks.
• To evaluate handcrafted versus deep-learning feature extraction methods.
• To assess the stability and real-time applicability of different models.

Main Methods:

• Image acquisition of six Iranian wheat cultivars under controlled conditions.
• Feature extraction using Principal Component Analysis (PCA) with Multi-Layer Perceptron (MLP).
• Deep feature learning with Convolutional Neural Networks (CNNs) using Global Average Pooling (GAP) and Fully Connected Layers (FCL) classifier heads.

Main Results:

• The CNN-GAP model achieved the highest accuracy (92.19%) and demonstrated superior generalization stability.
• PCA-enhanced MLP yielded 86.0% accuracy.
• Cross-domain testing revealed sensitivity to domain shifts, necessitating species-specific data.

Conclusions:

• The lightweight CNN-GAP architecture is suitable for real-time, low-cost deployment in precision agriculture.
• Automated classification significantly improves upon traditional methods for wheat seed identification.
• Further research should focus on domain adaptation for broader applicability.