Identifying defects and varieties of Malting Barley Kernels

Area of Science:

• Agricultural Engineering
• Computer Vision
• Machine Learning

Background:

• Accurate classification of malting barley kernels is crucial for quality assessment.
• Traditional methods for kernel analysis have limitations in precision and efficiency.

Purpose of the Study:

• To develop and evaluate a comprehensive approach for classifying malting barley kernels using dual-sided imaging and deep learning.
• To compare the performance of a custom-designed convolutional neural network (CNN) against traditional machine learning and transfer learning models.

Main Methods:

• Dual-sided kernel imaging and a custom image processing algorithm.
• Development of an optimized deep neural network architecture for barley kernel analysis.
• Comparative assessment of traditional feature engineering, transfer learning models, and the custom CNN.

Main Results:

• The custom deep learning network achieved 94% accuracy in classifying barley kernel defects and varieties.
• The proposed method outperformed established transfer learning models (93% accuracy) and traditional machine learning approaches (below 90% for defects, below 70% for varieties).
• Traditional methods showed an advantage in morphological feature recognition.

Conclusions:

• The developed deep learning network offers superior performance for malting barley kernel classification.
• Integrating morphological feature extraction with CNNs presents a promising direction for future research.
• Standardizing kernel orientation and merging dual-sided images are key for effective analysis.